Journal of Material Cycles and Waste Management

https://doi.org/10.1007/s10163-023-01794-6

REVIEW

A systematic review on smart waste biomass production using

machine learning and deep learning
Wei Peng1 · Omid Karimi Sadaghiani2

Received: 4 February 2023 / Accepted: 17 August 2023

© The Author(s), under exclusive licence to Springer Nature Japan KK, part of Springer Nature 2023

Abstract
The utilization of waste materials, as an energy resources, requires four main steps of production, pre-treatment, bio-refinery,
and upgrading. This work reviews Machine Learning applications in the waste biomass production step. By investigating
numerous related works, it is concluded that there is a considerable reviewing gap in the surveying and collecting the appli-
cations of Machine Learning in the waste biomass. To fill this gap with the current work, the kinds and resources of waste
biomass as well as the role of Machine Learning and Deep Learning in their development are reviewed. Moreover, the storage
and transportation of the wastes are surveyed followed by the application of Machine Learning and Deep Learning in these
areas. Summarily, after analysis of numerous papers, it is concluded that Machine Learning and Deep Learning are widely
utilized in waste biomass production areas to enhance the waste collecting quality and quality, improve the predictions,
diminish the losses, as well as increase storage and transformation conditions.

Keywords Machine learning · Deep learning · Waste biomass · Raw materials · Sustainable production

Abbreviations FRBS Fuzzy rule-based systems

ML Machine learning DBN Deep belief network
DL Deep learning CL Classification trees
ANN Artificial neural network C Carbon
IoT Internet of things O Oxygen
SVM Super vector machine S Sulfur
AAD Absolute average deviation A Ash
NB Naive Bayes K Potassium
KNN K-nearest neighbor P Phosphorus
DT Decision tree Ca Calcium
RF Random forest Zn Zink
ANFIS Adaptive network fuzzy inference system CO2 Carbon dioxide
XGBoost Extreme gradient boosting NIR Near infrared
GAM Generalized additive model RBF Radial-basis function
RNN Recurrent neural network ET Extra trees
MLR Multiple linear regression SPA Successive projection algorithm
RBNN Radial-basis neural network LRM Linear regression model
SMOR Sequential minimal optimization regression CRBM Conditional restricted Boltzmann machine
LDA Linear discriminant analysis GA Genetic algorithm
RO Reverse osmosis
* Omid Karimi Sadaghiani t-SNE T-distributed stochastic neighbor embedding
[email protected] DBSCAN Density-based spatial clustering of applica-
tions with noise
1
Faculty of Engineering and Applied Sciences, University GAN Generative adversial network
of Regina, Saskatchewan, Canada
GRU​ Gated recurrent units
2
Department of Energy Systems Engineering, Faculty SMR Stepwise multiple regression
of Engineering, Atilim University, Ankara, Turkey

13
Vol.:(0123456789)
 Journal of Material Cycles and Waste Management

LSTM Long short-term memory [1]. Basically, waste biomass is one the highly intended
CNN Convolutional neural network resource which can be found with low cost in a great vol-
MLP Multilayer perceptron ume. Primarily, the forest waste was used to warm the
FPN Mask Feature pyramid network mask home and cook the food or the urban waste can be used as
GP Gaussian process a fuel in thermal power plants. Like other forms of energy,
mAP Mean average precision the origin of existing energy in the waste materials is the
DNN Deep neural network sun and all of biomass raw materials provide their ener-
PR Polynomial regression gies from the solar energy. The utilization of biomass, as
GBDT Gradient boosting decision tree an energy resource, needs four main steps of production,
AdaBoost Adaptive boosting Pre-treatment, Bio-refinery, and Upgrading.
PLSDA Partial least square discriminant analysis Machine learning (ML), as a solver method of com-
RCCN Region-based CNN plicated engineering problems, is a combination field of
PGM Probabilistic graphical models computer science, approximate theory, probability theory,
GPR Gaussian processes regression and statistics [2]. With the use of machine learning (ML),
BNN Bayesian neural network which is a kind of artificial intelligence (AI), software pro-
LR Logistics regression grams may anticipate outcomes more accurately without
PLS-DA Partial least-squares discriminant analysis having to be explicitly instructed to do so. To forecast new
BRT Boosted regression tree output values, machine learning algorithms utilize past
GMMs Gaussian mixture models data as input [3].
LSSVR Least-squares support vector regression In supervised learning, data scientists describe the
GBM Generalized boosted model variables they want the algorithm to look for connec-
H Hydrogen tions between and provide the algorithms with labeled
N Nitrogen training data. The algorithm’s input and output are both
Cl Chlorine described. Unsupervised learning: Algorithms trained on
Pb Lead (plumbum) unlabeled data are used in this sort of machine learning.
Na Sodium The program searches through data sets in search of any
Mg Magnesium significant relationships. Both the input data that algo-
Si Silica rithms use to train and the predictions or suggestions they
HHV Higher heating value provide are predefined. Semi-supervised learning is a
PMF Positive matrix factorization method of machine learning that combines the two cat-
PLS Partial least squares egories mentioned above. An algorithm may be fed mostly
KRR Kernel ridge regression labeled training data by data scientists, but the algorithm
MARS Multivariate adaptive regression splines is allowed to explore the data on its own and come to its
CARS Competitive adaptive reweighted sampling own conclusions about the data set. Data scientists often
SVR Support vector regression use reinforcement learning to instruct a computer to carry
PCA Principal component analysis out a multi-step procedure for which there are set rules.
DO Dissolved oxygen An algorithm is programmed by data scientists to fulfill a
NF Nano-filtration goal, and they provide it with positive or negative feedback
LSA Latent semantic analysis as it determines how to do so. However, the algorithm
GNN Graph neural networks often chooses the course of action on its own.
GAT​ Graph attention network Deep learning is a machine learning method that instructs
LRLS Kernel-based regularized least squares computers to learn by doing what comes naturally to people.
GLM Generalized linear model Driverless vehicles use deep learning as a vital technology
to detect stop signs and tell a person from a lamppost apart.
It is crucial for voice control in electronic consumer goods.
Introduction Recently, deep learning has attracted a lot of interest, and for
good reason. It is producing outcomes that were previously
Waste-to-energy (WtE) is the process of turning trash unattainable. A computer model learns to carry out categori-
into a fuel source or producing energy from waste via the zation tasks directly from pictures, text, or voice using deep
primary treatment of garbage. The majority of WtE pro- learning. Modern precision may be attained by deep learn-
cesses either create a combustible fuel commodity, such ing models, sometimes even outperforming human ability.
as methane, methanol, ethanol, or synthetic fuels, or they A large collection of labeled data and multi-layered neural
directly generate electricity and/or heat via combustion network architectures are used to train models [4].

13
Journal of Material Cycles and Waste Management

This work investigates applications of ML in the waste and forest products. Therefore, the concept of the lignocel-
biomass production, because, by investigating a great num- lulose must be explained initially. Meanwhile, in addition
ber of related works, it is concluded that there is a consid- to agricultural and forest waste, some wilding and native
erable reviewing gap to collect the applications of ML in plants can also be used in bio-fuel production. For example,
the waste biomass origin, providing, monitoring, storage, Glycyrrhiza glabra L. (Liquorice) is an herbaceous perennial
and transportation. To fill this gap, in this review work, the legume native to Western Asia, North Africa, and Southern
resources of waste materials as well as the role of ML and Europe and can be converted to bio-fuel by Hydrothermal
DL in these areas are reviewed. Afterward, the storage and liquefaction (HTL) [5]. In other work, Durak worked on
transportation of waste materials are explained separately Trametes versicolor (L.) mushrooms or Verbascum Thapsus
followed by the applications of ML and DL in the develop- L. to produce bio-oil through HTL approach [6, 7].
ment of these areas.
Lignocellusic materials

Wastes biomass raw materials The lignocellulosic biomass is constructed by cellulose and
hemicellulose, and lignin. Lignin adheres to cellulose fibers
The waste types of biomass materials are provided by the to hold the plant cell walls.
waste and non-useable materials in farmlands or forests, Hemicelluloses are a chief part of the plant cell wall and
municipal wastes, industrial wastes, sludge, and livestock consequently impact on the extraction of cellulose and the
manures to be used as the resources of energy, as shown in quality of the wood and fiber. As a heteropolymer with side
Fig. 1. chains, hemicellulose is coupled with galactose, mannans, or
The wastes can be produced in different sections with alternate units of mannose and glucose (also known as glu-
various traits. Based on these traits, they can be collected comannans) or pentoses (xylans) (i.e., galactans) [8]. Vari-
and recycled to convert into useful and lucrative things like ous raw materials might include entirely different amounts
bio-fuels. Agricultural, Forest, Industrial, Municipal, and of sugar monomers. Cellulose is the highly existed organic
Animal wastes and Sludge. Also, lignocellulose is the main ingredient of Earth and the major structural part of walls in
constructor of the waste materials especially the agricultural biomass cells. Considering the chemical formula (­ C6H10O5)
n, cellulose refers to a long-chain polymer with a great grade
of polymerization and a high molecular weight. The poly-
mer known as lignin is produced by the oxidative coupling
of predominantly three monolignols, or p-hydroxycinnamyl
alcohols: p-coumaryl, coniferyl, and sinapyl alcohols. The
amounts of cellulose, hemicellulose, and lignin play a sig-
nificant role in progressing thermochemical processes in
biomass and, consequently, affecting the efficiency of the
herbal biomasses. However, experimental approaches for
the measurement of hemicellulose, cellulose, and lignin
amounts are high cost, time-consuming, and without eco-
nomic justification. More information about the structure of
the lignocellulosic materials can be accessed in [8].
Innovative methods for decomposing and converting
biomass may effectively dissolve the biomass, remove the
lignin from it, increase the output of sugars and lignin frag-
ments, and convert a large portion of it into fuels. The most
recent developments in lignocellulosic biomass valorisation
include optimizing lignocellulose pre-treatment, improving
saccharification and delignification through enzyme engi-
neering, and predicting the preferred genome modification to
transform products from biomass deconstruction into high-
value products [9].
The lignocellulosic biomass raw materials have been
focused by researchers to apply ML and DL to develop the
identification, production, and utilization of these materi-
Fig. 1  Categories of waste-origin biomass raw materials als. In one work, Kartal et al. [10] used ANN algorithm to

13
 Journal of Material Cycles and Waste Management

calculate the biomass fractions of hemicellulose, cellulose, < 8.1%, and RMSE < 0.8. Ionic liquids have been employed
and lignin by the results of proximate analysis, because the in wood production. RF technique is used by Phromphithak
experimental methods are expensive, time-consuming, and et al. [17] to predict the features of by Ionic liquid solvent
not commercially viable. In this work, the utilized ANN identities in the case that the prediction of features without
model gave the acceptable performance for all the major using ML is complicated. It was concluded that Random
biochemical components with the R2 > 0.96. Furthermore, forest algorithm was found to have the highest prediction
ANN, as an ML method has been employed to characterize accuracy. Sezer et al. [18] used ANN to predict the com-
the chemical exergy of different kinds of torrefied lignocel- bustion efficiency of woody biomass using its properties
lulosic fuels by Kartal et al. [11], because numerous actions and combustion conditions, so that his prediction can be
are required in experimental research. It was resulted that the used by policy makers and engineers, because thermogravi-
prediction capability of the ANN model was acceptable and metric analysis instruments are high-cost, and their appli-
the R2 was higher than 0.92 for training and 0.79 for testing. cation needs more time a trained operator. Eventually, the
The spectral data of samples also can be investigated to factors of MAPE R2 and RMSE for each biomass sample
categorize and compare woody biomass samples based on were surveyed, so that R2 > 0.99. In other works, a Bayesian
their lignocellulosic structures that have undergone vari- optimization-based ML framework has been used by Lof-
ous treatments and conditions, so that in 2021, Nag et al. gren et al. [19] to improve lignin solution, create high-value
[12] used the miscellaneous algorithms of RF, KNN, DT, bio-products, and raise the value of lignins by recognition
NB, XGBoost, and MLP for the categorizing the material of the optimal extraction conditions in lignin valorization.
and for the rapid analysis of biomass cell wall composition In summary, the applications of ML and DL in lignocel-
based on spectral analysis without a previous data of the lulosic structure of biomass raw materials are as: estimating
feedstock composition. The correlation between the amount structural fractions of lignocellulosic biomass, classifying
of produced bio-fuel and lignocellulose structure is another and comparing of biomass samples with varying types and
topic that Kogut et al. [13] in 2021 investigated by ANN species, predicting the bioethanol production amount, pre-
and RF. It was concluded that the hybrid model can be used dicting the conversion pathway of every biomass material
for the initial classification of plants only by their lignocel- relying temperature, time, moisture content, and structural
lulosic composition results and without the time-consuming components of lignocelluloses, estimating the heat capacity
and expensive research associated with pre-treatment and of the cellulosic biomass, examination of the impact of sev-
hydrolysis by the fit model of R2 = 0.961. In other work, eral process parameters, such as temperature, enzyme con-
Kardani et al. [14] focused on lignin, cellulose, and hemi- centration, biomass load, inoculum size, and time, predict-
cellulose parts of woody biomasses to estimate the conver- ing cellulose-rich materials properties from initial materials
sion in hydrothermal carbonization by XGBoost, MLP, and and requirements, such as biomass characteristics, operating
SVM without the need for any kinetic modeling. They pre- conditions, Ionic liquid solvents identities, and catalyst, pre-
sented an accurate model for the prediction of bioethanol dicting the combustion efficiency using biomass properties,
production versus lignocellulose elements. Accordingly, optimizing lignin bio-refineries, produce high-value bio-
XGBoost released a high-accurate results rather than the products and increase lignins value, and characterizing the
studied algorithms with R2 = 0.999 and 0.964 for training chemical exergy.
and testing data, respectively. Karimi et al. [15] focused Woody biomass analysis is one important step in achiev-
on estimating the heat capacity of woody biomass samples ing to the high-quality woody materials. The woody biomass
using LSSVR with the Gaussian kernel function which was can be divided into three groups of wood logs, chips, and
recognized as the best estimator. The validation of the uti- pellets. Wood as a biomass raw material has some advan-
lized model was carried out by 700 heat capacities of four tages over fossil fuel, including: (a) sustainability in supply
biomass samples (AARD = 0.32%, MSE = 1.88 × 10 − 3, of wood, as a renewable fuel, (b) lower emittance of ­CO2
and R2 = 0.999991). The total prediction precision of the from burning rather than fossil fuels, and (c) lower amount
LSSVR is higher than the empirical correlation. They pre- of sulfur and heavy metals. Besides, wood processing oper-
sented a model for analyzing the heat capacity based on lig- ations can produce wood waste, which includes sawdust,
nocellulosic parts. On the other hand, ANN, CT, and RF branches, bark, and leaves as well as needles.
approaches were utilized by Fischer et al. [16] to investigate Some researchers have been conducted around the woody
the impact of temperature, enzyme concentration, biomass biomass raw materials. For instance, in 2022, Kibleur et al.
load, inoculum size, and time on the generation of ethanol [20] studied the application of ML approaches and espe-
from lignocellulosic material during a simultaneous hydroly- cially DL in natural fiber composites and fiberboards which
sis and fermentation (SHF) without requiring the knowledge are considered as a woody biomass. The tissue of these
of the kinetics and the inhibition process. It was concluded composites is often intricate, and measurements of their
that R2 > 0.9 between actual and model predictions, AAD mechanical performance are incomplete. The utilized ML

13
Journal of Material Cycles and Waste Management

may facilitate the design of MDF, and helps the understand- height and above-ground biomass and accurately estimate
ing and improvement of wood-based materials. the kind of woody biomass and the forest structure, both of
To collect morphometric data from the fiber bundles which are important factors in the manufacture of biomass
and correctly estimate their density, CNN may be helpful. raw materials. The heartwood content of pine chips and bark
The pellet is measured by a precision caliper on a group content of birch chips were classified with an accuracy of
of individual pellets which needs more time and more rep- 91% using EIS from pure materials combined with a KNN
resentative samples. The geometric-dimensional properties classifier. On the other hand, 73% and 64% precise clas-
of wood pellets were determined using the Mask R-CNN sifications conducted using mixed materials and multiple
architecture DL in the previous study by Pierdicca et al. [21] classes for pine heartwood content for birch bark content,
in 2022. In several situations, including storage and feeding respectively.
circumstances, the length of the pellet is significant. The Utilizing tree height and crown diameter, ML tech-
density, energy density, and effective thermal capacity of niques are used to forecast each tree’s diameter at breast
thermal plants are all significantly influenced by the geomet- height which leads to an accurate estimation of carbon den-
ric-dimensional characteristics of pellets, which also have sity and forest structure. Malek et al. [26] employed SVM
an impact on flowability. The validation conducted by real and RF in this way by two datasets collected in southern
samples demonstrated high performances in terms of RMSE. Norway comprising 3970 and 9467 recorded trees, respec-
Plankenbühler et al. [22] provide an image-based approach tively. Accordingly, the developed ML techniques provided
for evaluating the quality of wood chips and mixing ratios. accurate results compared to a state-of-the-art work with
They discussed an online prediction approach based on RMSE = 81.4 kg and RMSE = 92.0 kg, respectively, for the
ML and concentrated on the fuel quality in biomass power prediction of AGB and DBH, respectively.
plants. Finding the mixture fraction from mixtures of two The light detection and ranging (LIDAR) technology has
wood chip species with varied quality and attributes is the been used to develop statistical models for identifying dif-
major objective. The gained models could predict with accu- ferent tree species and tracking the development of forests,
racy of R2 > 0.9 for the brightness-based model and R2 > 0.8 and the selection of regression method was controversial.
for the texture based. For this reason, Xue [27] fulfilled a thesis and used SVM,
ML also helps for rapid measurement of pellets and con- Cubist, and RF algorithms. Eventually, the training accura-
trolling the biomass quality during the production process. cies of 0.988 and 0.990, and validation accuracies of 0.902
Pitak et al. [23] manage the biomass quality throughout the and 0.853 were resulted for two study sites.
manufacturing process using quick measurements of pel- Therefore, the applications of ML and DL in woody
lets. Using color distribution mapping and line scan near biomass can be enlisted as follows: precisely estimation of
infrared-hyperspectral images, ML is used to estimate the fiber density, determination of the dimensional parameters
attributes of commercial biomass pellets, including the fuel of woody pellets, calculating the proportion of woods in
ratio (FR), volatile matter (VM), fixed carbon (FC), and ash mixes of two types of wood chips, prediction the proper-
content (A). The predicting model for FR, VM, FC, and ties of woody pellets (e.g., fuel ratio, volatile matter, fixed
A resulted in the R2 values of 0.75, 0.81, 0.82, and 0.87, carbon, ash content, etc.), predicting wood volume, above-
respectively. ground biomass, analyzing the bark and heartwood contents
To demonstrate DL-based 3D point cloud regression for of birch chips and pine chips, respectively, and predicting
estimating forest biomass, Oehmcke et al. [24] put together diameters at breast height of individual trees.
a study. Directly from 3D LiDAR point cloud data, DNNs
are used in their research to forecast wood volume, above- Agricultural wastes
ground biomass (AGB), and subsequently carbon. Wood
chip analysis may enhance the procedures for using these Agricultural wastes include the Rice straw, wheat straw,
products. Wood chip is well recognized as a raw material oat straw, barley straw, sorghum stubble, and corn stubble
for the pulp and bio-fuel industries. The DNN considerably (stalks, leaves, husks, and cobs). The collecting and selling
predicted wood volume, AGB, and carbon stock with higher of these wastes to a local bio-refinery leads to the increase of
accuracy rather than the state-of-the-art approaches. Con- farmers income beside the advantages of the production of
sequently, ML and remote sensing can be used to perform renewable bio-fuels such as positive environmental impacts
these measurements of vegetation structure at large scale. [28].
The heartwood content of pine chips and the bark content of In some cases, ML and DL approaches have been used
birch chips were examined by Tiitta et al. [25] using electri- in the field of Agricultural wastes. For an outstanding sam-
cal impedance spectroscopy (EIS) in conjunction with ML ple, Xue et al. [29] presented a quick and simple way using
in 2020. Three ML techniques (i.e., KNN, DT, and SVM) image-based DL to forecast the maturity of compost made
were used to analyze the distribution of diameters at breast from agricultural waste. Unquestionably, compost maturity

13
 Journal of Material Cycles and Waste Management

is a crucial metric for assessing the efficacy of compost- stumps, and (ii) developing the efficient tools to identify
derived goods. Biochemical tests as conventional methods wood species accurately.
for investigating compost maturity are time-consuming and The process of wood monitoring, grabbing, holding, and
difficult to perform. For this reason, examining photos of harvesting is a very important factor in the sustainable pro-
various composting phases and applying the CNN technique, duction of wood biomass. According to the monitoring, a
the maturity of the compost is investigated rapidly. The forests cover minimum 0.5 ha and canopy cover exceeds
accuracy of proposed method was more than 99% for 4 test 10% and the height of trees is higher than 5 m. Thus, around
sets. Thus, the applications of ML and DL in the agricultural 33% of the earth surface is forests, of which 26% are pri-
waste biomass field are presented as: predicting agricultural mary forests and about 5% of the forests have been planted
waste compost maturity. exclusively for woody biomass production. In recent years,
more than 2 billion people require wood for energy supply.
Forest wastes Approximately, a great portion of the total biomass is pro-
vided by forests all over the world.
Basically, the wastes produced in natural forest and the pro- A continuous group of trees that are uniform in composi-
duced wood of planted forests are two main sources of the tion, structure, age, size, class, distribution, spatial arrange-
production of woody biomass materials and are investigated ment, site quality, condition, or location are referred to as
followed by the applications of ML and DL. a forest stand. This helps to differentiate it from nearby
Natural forests, as the main producer of biomass, is groups. The following five factors describe the make-up of
known as a terrestrial ecosystem and its produced biomass commercial forest stands are as follows: (i) Uniformity of
is varied depending on location, stand composition, geogra- stands species, (ii) Structure of canopy density (horizontal
phy, weather, operational procedures, and disruptions [30]. or vertical), (iii) Variety of species in a stand, (iv) Stand
Generally, in all forests, thinning and pruning may produce composition from miscellaneous species, and (v) Stocking
biomass or mass arboriculture with the aim of harvesting system (difference in species or difference in size).
the timber and energy crop as well as grabbing forest resi- Basically, forest growth is measured and determined by
dues removed from the stands. A considerable volume of two indices of Tree growth and Stand growth. Tree growth
these products can be used as biomass raw material. Due to is defined as the augmentation of the dimension and mass,
low density of forest residues and being some transporta- and stand growth is specified by the stem diameter and form,
tion problems, the costs augment, and thus, it is lucrative to tree height, and crown area.
attenuate the biomass density in the forest area. Sustainable forest management (SFM) is known as “The
Automated mapping of woody waste over harvested for- supervision and application of forest materials in a non-
est plantations utilizing UAVs, high-resolution photography, destructive way without threatening their natural produc-
and ML was the subject of study by Windrim et al. 2019 tivity, ecosystem, biological cycle and their capability to
[31], because traditional methods for automated mapping conduct some operations at different levels”. Thus, seven
of woody waste are time-consuming. An essential stage in indices have been mentioned in the SFM presented by the
controlling the extraction of residues, monitoring site qual- International Tropical Timber Association (ITTO). These
ity, and resolving discrepancies between pre-harvest inven- ecologic, economic, and socioeconomic indices are: (i)
tory and actual timber yields is the investigation of woody Production of facilities for SFM, (ii) Extent and enhance
debris left over following harvesting operations on managed living condition of forests, (iii) Forest ecosystem hygiene,
forests. Modern machine learning and image processing (iv) Species distribution models, (v) Forest production rate,
methods were used to detect, segment, and count both fine (vi) Hazard assessment, (vii) Forest biological diversity,
and coarse woody debris, including tree stumps. Finally, (viii) Forest carbon cycles, (ix) Soil and water conservation,
automated algorithms could detect stumps with a precision and (x) Economic issues. The study of SFM has recently
of 83.9% and a recall of 81.8%. increased.
He et al. [32] examined the creation of effective tools for The structures of natural fiber composites and fiber-
precisely identifying wood species in his prior studies. They boards are complicated, and their mechanical behavior is
applied four ML classifiers of DT, NB, SVM, and ANN not yet fully understood. For this reason, Kibleur et al. [20]
to discriminate between the species. Eventually, SVM gave in 2022 utilized ANN, GBM, and RF to focus on a desired
the best performance among the mentioned algorithms, with composition and age structure or forest biological diversity
an overall accuracy of 91.4% and a per-species R2 = 66.7%, leading to sustainable forest management. Understanding
95.0%, and 80.0% for S. humilis, S. macrophylla, and S. and recognition of the forest condition after forest fighting is
mahagoni, respectively. Therefore, the applications of ML an important step in the production of forest wastes. Aerial
and DL in forest wastes are as follows: (i) Finding, classify- image-based forest analysis can give an early detection of
ing, and counting coarse and fine woody waste, such as tree dead trees and living trees. After a forest fire, dead and live

13
Journal of Material Cycles and Waste Management

trees may be identified early using aerial image-based forest by the RF approach have much more precision and distinc-
analysis. Automated dead tree recognition from aerial pic- tion ability than the LR and NN models in all conditions
tures is done using DL technique. For instance, Chiang et al. leading high accuracy (> 75%) and good distinction ability
[33] used the algorithm of Mask R-CNN in the recognition (AUC > 0.8). Also, the likely zones for deforestation have
of the woody unfired residuals by mean average precision been determined using ML algorithms. In one prominent
score (mAP) of 54%. work, Saha et al. [40] analyzed the deforestation areas by
Furthermore, in Brazil’s eucalyptus plantations, ML SVM, NB, RF, DT, and ANN. Tracking land usage for
approaches have been employed to forecast the amount of soil and water conservation is another application of DL
wood with good results. The algorithms of ANFIS and RF in the forest monitoring process. Accordingly, SVM gave
have been applied by Silva et al. [34] in the prediction of more accurate results rather than other models validated via
woods. The data were collected by continuous forest inven- receiver operation characteristics, efficiency, and sensitivity
tories conducted in stands of eucalyptus clones. ANNs and and specificity measurement. The AUC amounts of these
DT were used to compare the feasibility of the models. The models are 0.907, 0.885, 0.825, 0.846, and 0.876, respec-
evaluation of the estimates generated by the techniques for tively. Wan et al. [41] used the Deep CNN to continuously
training and validation was calculated based on the follow- analyze the change in soil coverage. This model can easily
ing statistics: r, RB, and the RRMSE. Basically, the utilized discriminate edge features from remote-sensing images by
techniques gave the best performance for the training and high pixel accuracy of 83.4%.
validation data set so that the amount of RRMSE factor The management of mechanical timber harvesting and the
is lower than 6%, RB is lower than 0.5%, and r is higher production of sustainable woody biomass may both benefit
than 0.98. Besides, the SC and FCM algorithms provide from the information on forest activities. As part of a forest
precise baseline and volume projection estimates and can operation, ET, RF, and DT algorithms are used to estimate
be considered as logical alternatives for choosing the effec- the productivity of wood harvesters by Munis et al. [42]. The
tive parameters applied in modeling of the forest production. applied dataset was consist of more than 144,000 files pro-
Yousafzai et al. [35] utilized RF and KRR mode to fore- vided during 28 months. They analyzed the predictive per-
cast forest yield, using temperature, humidity, rainfall, and formance of 24 ML algorithms and some combined models.
wind speed as the predictive factors under various climate The accuracy was checked by the RMSE, MAE, MAPE, and
change scenarios. Prediction accuracy of both the models R2. Learning by combined models ended to R2 = 0.71 and
was evaluated by means of RMSE, MAE, r, RRMSE, LM, MAPE = 15%. Lee et al. [43] used Plot-based observations
WI, and NSE metrics. Overall, the RF model outperformed and airborne LiDAR data to use ML to create stand height
the KRR model due to its higher accuracy in forecasting of estimate models for forest inventory data. They applied
forest yield. Meanwhile, the social and ecological effects of SVR, RT, and RF algorithms for effective sustainable forest
forest and other natural resource policy have been surveyed management over large regions. Eventually, no considerable
by Rana et al. [36]. They used CT technique to evaluate the difference was reported among the algorithms in estimation
effects of forests. Precise volume growth estimates of forest of heights and bias (p values > 0.05). In other work, RF and
ecosystems are necessary to comprehend carbon sequestra- PMF have been used by Zhang et al. [44] to assign sources
tion and reach carbon neutrality objectives. By simulating of heavy metals as soil contaminants and forecast their geo-
the evolution of the forest stand, ML is used to estimate the graphical distributions. Their studies focused on coastal
volume growth of Larix and Quercus forests. For this pur- reclaimed areas. Accordingly, the RF algorithm performed
pose, Tian et al. [37] applied RF method in the estimation of better in predicting heavy metal distributions by surveying
forest volume growth. According to the results, the stand age soil variables. In this work, the performance of RF model
can be considered as the main predictor for volume growth was investigated by two parameters of MSE and Var, while
and the performances of the RF model were roughly stable the applied error indices for evaluating the performance are
(R2 = 0.40–0.66). ME, MSE, RMSE, and R2.
Zhang et al. [38] used CNN and Keras to reach a carbon Accurately mapping the forest vegetation in riparian
peak and attain carbon neutrality by accurate evaluation of zones is a crucial problem in the protection of the quality of
forest carbon storage. It was concluded that the CNN model water supplies. To help the process of mapping vegetative
performed better than traditional models and results of for- regions, information from remote-sensing data is extracted
est carbon-density estimation verified the models by merg- using ML methods like DT, RF, SVM, and NB by Furuya
ing the optical and radar data. For a precise assessment of et al. [45]. It was concluded that DT learner has the highest
regional-scale forest above-ground carbon density, DL has accuracy in forest vegetation mapping on different dates and
been utilized. ANN and RF algorithms have been applied river sides. On the other hand, according to the experience
by Hart et al. [39] to forecast which specific trees in a forest of Chen et al. [46], SVM and RF can help forest distribution
are most vulnerable to storm damage. The models provided and forest damage estimation. Moreover, a novel non-linear

13
 Journal of Material Cycles and Waste Management

model of the above-ground carbon stock in a secondary automatically into some categories like organic, inorganic,
semideciduous forest was proposed via SVM and ANN by and medical waste, presenting an automatically classifying
Dantas et al. in 2021 [47]. Their applied algorithms released system of garbage, picking garbage on the grass efficiently
the highest accuracy results of OA = 80.36% in both cases, and automatically, predicting the amount residues and waste
and ROC = 0.89, AUC = 0.88, respectively, by Fractional materials of destroyed buildings, development of the fore-
Cover as a predictor variable. Among the vegetation dam- casting household waste generation, reduction of huge labors
age indices, DNDVI resulted the highest accuracy, so that of recycling tasks, detection, classifying, and analyzing the
AUC = 0.85 and OA = 77.68%. various types of street litters, classifying plastic waste into
To summarized, the application of ML and DL in SFM some categories based on their compositions, predicting
can be enlisted as: focusing on a desired composition and the amount of air pollution in the vicinity of the garbage
age structure or forest biological diversity, early detection environment, simultaneously identification of MSW and
of dead trees and living trees after forest firing, predicting its location in images, detection and classifying the differ-
the volume of wood, forecasting the raw material production ent pollutants, creation of innovative battery extraction and
of forest during different climate conditions, analyzing the sorting processes, predicting higher and lower heating value
social and ecological impacts on forest, acquiring knowledge of MSW, improving the MSW circularity, and predicting
about carbon sequestration and being carbon neutral, accu- MSW’s hot flue gas temperature and syngas heating value
rate estimating of regional-scale forest above-ground carbon at a waste-to-energy facility.
density, predicting the individual trees risk of damage after
storm, identifying the deforestation probable zones, survey- Industrial wastes
ing land usage to ensure the preservation of soil and water,
estimating the productivity of timber harvesters, as a forest Industrial wastes can contribute considerably in providing
operation information, developing stand height estimation, biomass raw materials and Food industry produces the major
apportioning sources of heavy metals as soil contamination part of industrial wastes. The waste part of food is a world-
and predict their spatial distributions, estimating forest dis- wide topic that has recently fascinated public and govern-
tribution and forest damage, extracting information from ments attention. Nearly 1300 million tons (i.e., 33% of food
remote-sensing imagery, and proposing a new non-linear provided for human nutrition) as well as a great volume of
model of the forest's above-ground carbon stock. water (nearly 250 million m3) is lost annually [68, 69]. Thus,
food waste effects economically and environmentally cli-
Municipal solid wastes mate change and governments, companies, and researchers
have tried to diminish the wastes or recycle them for provid-
Municipal solid waste (MSW) is produced from garbage ing biomass or livestock feeding [70, 71].
generated in homes, businesses, parks, and leisure areas. Waste from all sectors of the food industry can be consid-
Thus, it is composed of organic compounds and wood and ered as an energy source, such as waste water after washing
directly depends on the population of a city. Roughly 80% meat, fruits and vegetables, etc. These effluents have a great
of MSW consists of containers, paper, plastics, and food deal of sugar, starch, and other soluble and solid organic
wastes. MSW can be directly combusted to be used or con- matter and are converted to bio-fuels by anaerobic digestion
verted to energy through natural anaerobic digestion hap- or fermentation.
pening at landfills where natural decomposition of MSW Similar to other types of wastes, industrial wastes have
produces landfill gas (LFG) including nearly 50% C ­ H4 and also been analyzed with ML and DL approaches. By study-
50% ­CO2. The generated methane is utilized to generate ing numerous works, the application of ML and DL in the
electricity in power plants or internal combustion engines field of industrial waste was investigated. For example,
[48]. Smart cities have a number of waste management chal- Alberto Garre et al. [72] worked on application of ML to
lenges, such as inadequate MSW knowledge, incorrect MSW aid in the planning of output in the food business to moni-
categorization, organized sector engagement for effective tor and diminish the wastes. A food company that produces
MSW management, a lack of technology, and transporta- liquid products was investigated and data of 1795 batches
tion of MSW. were gathered. In their study, ML algorithms were utilized
After analyzing the conducted research, the applications to predict variation in production and decrease the uncer-
of ML and DL in MSW area are presented in Table 1. tainties associated with the value of waste in access. It
In compliance with Table 1, the applications of ML and was concluded that these accurate predictive models can
DL in MSW field include: analyzing the waste index of a decrease their environmental effects. Sławomir Czarnecki
certain disposal site, recognition of MSW content, facilitat- et al. [73] applied DT, RF, and AdaBoost algorithms for
ing equipment selection for MSW management, enhance- the accurate recycling of the waste granite powder manu-
ment of MSW classification accuracy, classification of MSW facturing of green cementitious composites modified. They

13
Journal of Material Cycles and Waste Management

Table 1  Applications of ML and DL in MSW

Author Application Algorithm

Khan et al. [49] The dustbin’s contents are dispersed around the streets as a result of inef- SVM, RF, MP, and NB
ficient rubbish collection. Because of this, it is suggested to look into the
trash index of a specific dump site using ML and the Internet of Things as a
practical option for smart and effective garbage management efficiently and
effectively, because smart bins, collection vehicles, and routes are dynami-
cally updated
Sevcik et al. [50] Near-infrared, hyperspectral imaging, and ML methods are used to enable PLSDA, SVM, and
classification of typical MSW components by classification of pixels RBNN
belonging to textile, wood, food, paper, and most common types of plastics
with reasonable accuracy. The overall accuracy of SVM model calculated
from classification sensitivity was 85% in prediction of external sample set
The accuracy of RBNN model was 85% in prediction
Minimum computational time and lowest prediction accuracy belong to PLS-
DA
Ali et al. [51] Making a decision needs biased decisions, time-consuming analyses, and an LR, MLP, DT, and SMOR
unpredictable future. ML is used to facilitate equipment selection for the
MSW management and demonstrate the applicability of decision-making
framework
Huang et al. [52] The prior waste classification models using DL were not easy to achieve CNN models (VGG19, DenseNet169, and
accurate results and they should be improved due to the various architecture NASNetLarge)
networks adopted. Since different kinds of wastes need different disposal
methods, the precise and dependable categorization technique is a crucial
element in the waste management process. DL with existing waste clas-
sification models improved by the various architecture networks can create
high waste classification accuracy of 96.5% and 94%
Vo et al. [53] Rapid and automatic smart waste sorter can be done by ML. A strong DL DNN-TC: an improved ResNext
method is proposed to classify wastes automatically into three categories of
organic, inorganic, and medical waste
DNN-TC resulted 94% and 98% accuracy for Trashnet and VN-trash datasets,
respectively. Also, DNN outperforms the state-of-the-art methods
Kang et al. [54] Garbage classification can be considered as an essential parameter of ResNet-34
resource conservation, social sustainability, and environmental preserva-
tion. To improve the performance of garbage collection, DL is used to pre-
sent an automatic garbage classification system. The classification accuracy
of ResNet-34 increased up to 99% and the classification cycle speed of the
system is 0.95 s
Bai et al. [55] The autonomous navigation and garbage identification features are presented SegNet and ResNet
using a unique navigation technique in DL, so that the robot can pick
garbage on the grass efficiently and automatically. The garbage recognition
accuracy can reach as high as 95%
Kumar et al. [56] To develop smart cities, smart waste management system is required which YOLOv3
can be done by the ML techniques. In some areas, building materials may
have biomass resources. Due to the lack of time, planning for the reuse
of construction materials at the end of a building's useful life is often a
challenging issue. To solve this problem, DL models are developed for pre-
dicting the amount of residues and waste materials of destroyed buildings.
YOLOv3 was suitable method for all the classes of waste items
Cubillos [57] Due to its great unpredictability and uncertainty, forecasting household Multi-site LSTM CNN
garbage output using conventional approaches is extremely difficult. Com-
pared to municipal or national levels, household data might exhibit faster
short-term fluctuations and stronger non-linear dynamics. The forecast-
ing of home trash production is created using a cutting-edge DL method.
Eventually, it was concluded that the multi-site LSTM model outperforms
the traditional models by 85% on average
Mao et al. [58] DL was required to improve the traditional image recognition technology. CNN-DenseNet121
Therefore, DL may assist to cut down on a lot of labor-intensive recycling
duties. The data augmentation strategy may increase the waste classifica-
tion’s accuracy and a genetic algorithm (GA) can be adopted. In this work,
TrashNet with 2527 images with six different waste categories was applied
to assess the CNNs’ performance

13
 Journal of Material Cycles and Waste Management

Table 1  (continued)
Author Application Algorithm

Ping et al. [59] Conventional street cleaning methods need a considerable amount of manual Fast region CNN
operations for detection and evaluation of street’s cleanliness which leads
to a high cost. A DL model, as strong potential and effectiveness in smart
city systems, is developed to find, sort, and analyze different kinds of street
trash, including bottles, leaves, and tree branches
Bobulski et al. [60] The separation of renewable waste is a big challenge for many countries. The CNN
classifications of polyethylene terephthalate, high-density polyethylene,
polypropylene, and polystyrene are described together with a DL technique
for categorizing plastic waste in these materials
Hussain et al. [61] The air pollution concentration in the vicinity of the bin environment is Simple and modified LSTM CNN
predicted using ML and DL models. The system caused a 4 s delay between
when the alarm message was created and sent to a sanitary worker. The
accuracy amounts of modified and simple LSTM models are 90% and 88%,
respectively, to predict the future concentration of gases found in the air
Liang et al. [62] Evaluating the multi-label waste classification and localization tasks to Multi-task learning architecture-based CNN
advance waste management was a necessary action conducted by DL. A
multi-task learning architecture (MTLA) relying on a DL model is utilized
to at the same time identify and localize wastes in images. The average
precision score was over 81.50%
Panwar et al. [63] Every year, more than 8 million tons of plastic waste are poured into the seas. RetinaNet
The suggested DL model finds and categorizes the many pollutants and
hazardous waste materials floating in the water and along the coast. The
grabbed plastics can be used to convert into oil using pyrolysis conversion
method
AquaTrash was the used dataset based on TACO data set. Also, the proposed
model detected and classified the waste items in the oceans and lake banks
with mAP of 0.8148
Sterkens et al. [64] Prior studies about automatic waste detection were hardly comparable YOLOv2
because of the lack of benchmarks and standards. Innovative methods to
sort and extract batteries from waste electric and electronic equipment are
becoming more and more necessary. Thus, by assessing the technological
viability of anticipating the existence, position, and type of batteries within
electronic devices, the DL approach is utilized to promote the development
of innovative procedures for battery extraction and sorting. The presented
approach resulted up to 70% of average accuracy in waste detection and
near 75% of classification accuracy on the test dataset
Birgen et al. [65] An ML model can predict LHV of MSW which would be a great step to GPR/ML
improve operation of plants, and their overall performance. The training
dataset consisted of 730 observation points and the GPR algorithm accu-
rately was performed and learned from new data
Bijos et al. [66] The composition, production, and the utilization rate of MSW, the consump- SVM and RF
tion patterns change, and social contribution of informal waste gatherers
have developed the concept of circularity of MSW. Improvements in MSW
circularity can be accomplished using ML models. In this work, Web of
­Science® database, the B­ iblioshiny®, and the ­Bibiometrix® package were
used
ML methods could proliferate income generation for informal waste collec-
tors in Latin America and Caribbean
Kabugo et al. [67] By combining cutting-edge ML techniques with big data processing tools and Neural network-based NARX
cloud computing technologies, data-driven soft sensors that can be used to
monitor a waste-to-energy plant may be developed. In a waste-to-energy
facility, data-driven soft sensors were investigated to forecast the syngas
heating value and the hot flue gas temperature of MSW
The NN-based NARX model released more precise performance in the pre-
diction of both syngas heating value and flue gas temperature

13
Journal of Material Cycles and Waste Management

applied a database consists of 216 sets of data provided by they employed the DBN framework. Because mine sludge
parameters, such as the percentage of cement substituted deposits have a significant potential environmental impact,
with granite powder, and time of testing and curing condi- M. Araujo et al. [80] put together a paper titled "Use of
tions. Accordingly, the utilized ML technique for designing machine learning techniques to analyze the risk associated
green cementitious composites without producing waste has with mine sludge deposits" in which MLP neuronal networks
more accuracy than traditional tests performed in a labora- are used to estimate the risk of the sludge deposits. ML and
tory is essential. DL were therefore used in the sludge sector for: describing
Sorting the wastes is another application of ML and DL emissions such as ammonia, microplastics, etc., monitoring
in industrial waste scope which was conducted by Sishir the sludge profile, classifying the early maturity of compost,
Bhandari [74]. This researcher utilized SVM and CNN algo- sludge bulking monitoring, and estimating risk of the sludge
rithms to automatically sort the wastes in industrial centers. deposits.
Therefore, ML and DL techniques can be applicable in waste
quality monitoring and predicting, diminishing the wastes,
Animal wastes (manure and livestock wastes)
accurately recycling the waste, and automatically sorting the
wastes.
One of the main raw materials utilized as a biomass energy
resource is livestock and manure which can be converted into
Sludge
bio-fuel (bio-gas) through anaerobic digestion method. Also,
sewage, as an animal waste, can be anaerobically digested to
Sludge and organically polluted wastewater can be regarded
the bio-gas. Then, the left sewage sludge can be combusted or
as main resources of biomass. These resources may be pro-
converted into bio-gas by pyrolysis process [81].
duced by households, municipal facilities (e.g., military,
In compliance with the studied works, there are some
academic, and clinic centers), or small companies (e.g.,
applications of ML ad DL approaches in animal waste scope.
laundries and restaurants). Thus, relying on the place of
For instance, using machine learning as a forecasting tool,
production, sludge can be considered as both industrial and
Ihsan Pence et al. [82] put together a report titled thorough
municipal wastes. The extraction of bio-fuel from sludge can
analysis of animal manure-based C ­ O2 emissions, coal, elec-
be realized by aerobic and anaerobic fermentation [75]. ML
tricity, thermal energy, and CH4 emissions. SVM, MLP,
and DL approaches have been used in the field of sludge. ML
and LR algorithms were used in this study. In a different
and DL approaches have been used in the field of sludge.
research, Stephen G. Matthews et al. [83] introduced auto-
The related study by Jess Zambrano et al. [76] focuses on
mated tracking to monitor pigs’ wellbeing and follow behav-
Machine Learning approaches for monitoring the sludge pro-
ioral changes. For the purpose of monitoring pigs’ changing
file and applies two ML methods for monitoring the sludge
behavior, they used GMMs. By utilizing SVM classification,
profile in a secondary settler tank utilizing GPR and GMMs’
Jaime Alonso et al. [84] enhanced the estimate of bovine
methods. According to the results, GMM provided a faster
weight trajectories. One of the uses for ML is the prediction
response with more sensitivity rather than GPR. Based on
of cattle weight trajectories for future evolution using only
pH and the carbon-to-nitrogen ratio, P. Boniecki et al. [77]
one or a few weights. Thus, the applicants of ML and DL in
use ANN to characterize ammonia emissions from com-
the scope of animal waste can be enlisted as estimating of
posted sewage sludge in additional studies. Eventually, for
cattle weight, examination of C ­ O2 emissions resulting from
all of the selected models, the correlation coefficient reached
animal manure, and behavior change monitoring.
the high values of 0.972–0.981. Sebastian Kujawa et al.
[78] considered grading the maturity of compost based on
sewage sludge and rapeseed straw in addition to evaluating Waste storage
ammonia emission. Since one of the most well-liked and
economical ways to handle sludge is via composting. The Basically one of main traits of biomass raw materials is their
early maturity of compost created from sewage sludge may seasonal availability and limited harvesting time within a
be categorized by CNN. These composts are crucial to the year which effects on crop sustainability. Meanwhile, a
creation of biomass. 25 CNNs with a great deal of filters sustainable supply of bio-fuel usually depends on storage
were developed and the classification error of the developed facility and techniques. The storage techniques can be actu-
models ranged from 0.51 to 17.77%. A significant challenge ally categorized as: (a) Open-air storage without protection,
in determining the amount of sludge bulking is monitor- (b) Covered storage without climate control, (c) Ensilage
ing the sludge volume index. Safder et al. [79] calculate in bags, (d) Covered storage with the potential for climate
the sludge volume index and categorize the bulking states control, and (e) Silo used for pellet storage [85].
using a hybrid DL-based soft sensor. For sludge bulking
monitoring in a dynamic wastewater treatment process,

13
 Journal of Material Cycles and Waste Management

Biomass open-air storage method is the low-cost method in silos employing knocking sounds was introduced by
and is usually carried out on-field or in the proximity of Paaranan Sivasothy et al. [91]. The SVM approach is used
the bioenergy instruments. Having the maximum loss and in this study to categorize various filling levels using various
quality declination values, lack of the moisture controlling, data patterns.
self-ignition, and long-date storage are considered as disad- In other research, Lorenzo Serva et al. [92] evaluated
vantages of this method [86]. the rise in acetic acid production in silage, because strains
In a covered-without-climate-control storage method, of Lactobacillus buchneri promote acetic acid synthesis in
miscellaneous kinds of biomass covering are applied to silage, suppressing yeasts and enhancing aerobic stability.
diminish the devastating impacts of weather on loss and Therefore, ML is used to study the likelihood that the start
quality of dried biomass. This kind of storage provides of aerobic instability is connected to a number of variables,
greater protection from snowfall or rain. Although more including the use of various Lactobacillus buchneri doses,
expensive, covered structures with walls may shield biomass the gradual delaying of silo closing, and the nutritional com-
from wind and tornadoes [87]. When utilized for anaerobic position at harvest. To proactively monitor the silo bag-kept
digestion by bacteria that ferment carbohydrates to create grain condition, such as temperature and moisture, Shawqi
lactic acid, ensilage of biomass is an acceptable approach Mohammed et al. [93] use a super vector machine classifier
for wet storage of high-moisture biomass [88]. With the of grain stored in silo-bags as well as a wireless network of
least amount of dry matter losses, the ensilaging process temperature and moisture sensors.
preserves biomass throughout time. The cover with tempera- In another study, Alina Kowalczyk-Juko et al. [94]
ture control storage technique is comparable to the ambi- investigated how to quantify the methane generation from
ent moisture control system fitted biomass ensilage storage different substrates in the form of silages by utilizing the
method. Three key characteristics of this storage approach energy value of the silages. In a different work, Huseyin
are maximum cost, minimal material loss, and control over Duysak et al. [95] employed a KNN algorithm to estimate
biomass moisture levels. In particular, sloping floor silos the amount of grain in a silo using air radar. The provided
may be used for pellets or woodchip. The conversion and accuracy was 98–99% in automatic counts in comparison
transportation of biomass are made easier by this technol- to actual transportation kinds. Éverton Lutz et al. [96] has
ogy, although the biomass must first be processed into either carried out a thorough investigation into the applicability of
chopped or pelletized form. new technologies for tracking and forecasting grain storage
Numerous works were surveyed in detail which utilized quality and amount using ML and IoT. To summarize the
ML and DL algorithms in the biomass storage scope. Using application items of ML and DL in biomass storage, the fol-
LR, DT, RF, and MLP, for instance, Diana Goettsch et al. lowing list is presented as: choosing viable sites for storage
[89] were able to choose probable storage depot sites and depots and improving the solution, predicting under-treat-
enhance the solution. According to the results, the MLP is ment paddy moisture content, classifying different filling
shown to be the best and most acceptable method for choos- levels in silage, investigating the risk of aerobic instability
ing suitable depots, resulting in a drop in the overall cost of onset, proactively monitoring the silo bag-stored biomass
wood storage and sustainable biomass processing. On the condition such as temperature and moisture, determining
other hand, the LR and the DT well performed in the cost grain quantity in silo via air radar, estimating the methane
reduction. production, and monitoring and predicting quality and quan-
The MLP is shown to be the best and most acceptable tity of stored biomass.
method for choosing suitable depots, resulting in a drop in
the overall cost of biomass storage and sustainable biomass
processing. Waste transportation
Since hot air drying in a storage silo is an alternate
method to paddy drying, Farooq Ahmad et al. [90] con- Transportation is an important issue in the biomass produc-
ducted research on ML-based grain moisture estimate via tion step which plays an axial role in the stable production
supervised learning approaches. By periodically recondi- of biomass. It is defined as the displacement of biomass raw
tioning stored grains over a longer storage time by push- materials from farmland to bio-refinery spots [97]. Miscella-
ing hot air over grains, it may assist to preserve moisture neous forms of transportation can be utilized, such as truck,
content in safe limits. In light of historical trends, internal train, pipeline, barge, or a combination of them.
environmental data, and external environmental parameters, Like the prior scopes, the transportation has been devel-
an ML approach is created to forecast under-treatment paddy oped by ML and DL. However some of these works are
moisture content. The bulk solid silo's fill level should be not exclusively for biomass raw materials, they can be
understood, since it is crucial to the biomass storage pro- applicable for these materials. For instance, using data
cess. A new acoustic sensor method for fill level assessment from January 2010 to December 2016, Milenkovi et al.

13
Journal of Material Cycles and Waste Management

[98] used an ANN with simulated annealing and a genetic asset maintenance, forecasting the demand, path way and
algorithm to estimate the demand for trailers in the port on-time performance prediction, truck traffic prediction, and
of Barcelona. Using data from the previous month’s con- behavioral analysis and facility planning.
tainer throughput at Singapore and Shanghai seaports, Niu
et al. [99] estimated container throughput using SVM. In
Conclusion
another study, Loaiza et al. [100] concentrated on air-
planes and utilized ANN and linear regression to estimate
This review paper has been compiled after precise study-
the demand for air cargo using several inputs, including
ing and analyzing numerous papers around the applica-
the rate paid for the carrying of cargo, GDP, the kinds of
tion of ML and DL in the waste biomass production. Since
items, and the area that was used to create the products.
the authors tried to cover all accredited papers and theses,
In the intermodal domain, Moscoso-López et al. used
approximately the majority of areas and scopes have been
ANN and SVM algorithms to estimate the daily Ro–Ro
collected. Beside the analysis of the application area, the
vegetable freight volume by analyzing the port of Alge-
utilized ML or DL algorithm was surveyed. After analysis
ciras Bay using Ro–Ro vegetable freight data from previ-
of numerous papers, it is concluded that Machine Learn-
ous days. Other applications of ML and DL in biomass
ing and Deep Learning are widely utilized in waste biomass
transportation include operation and asset maintenance in
production areas to enhance the waste collecting quality and
addition to demand forecasting. For instance, Du et al.’s
quality, improve the predictions, diminish the losses, as well
study [101] on vessel routes across the world focused on
as increase storage and transformation conditions.
predicting fuel consumption rates for the vessels utilizing
input characteristics, such as sailing speed, displacement,
trim, weather, and sea conditions, along with an ANN References
technique. Gao et al. [102] used the LSTM technique in
1. Anjum M, Umar MS (2018) Garbage localization based on
terminal operation to predict the daily container volume
weakly supervised learning in Deep Convolutional Neural Net-
of Chinese ports using the container volume data from the work. In: Proceedings - IEEE 2018 International Conference on
previous year. Shyur et al. [103] evaluated the airport of Advances in Computing, Communication Control and Network-
USA taking into account age, expected flight hours, and ing, ICACCCN 2018. pp 1108–1113. https://​doi.​org/​10.​1109/​
ICACC​CN.​2018.​87485​68.
projected landing time to determine component inspec-
2. Thimm O, Bläsing O, Gibon Y, Nagel A, Meyer S, Krüger P et al
tion needs for asset maintenance. Path way and on-time (2004) MAPMAN: a user-driven tool to display genomics data
performance prediction is the third branch of transporta- sets onto diagrams of metabolic pathways and other biological
tion that ML and DL have established. Through the use of processes. Plant J 37(6):914–939. https://d​ oi.o​ rg/1​ 0.1​ 111/j.1​ 365-​
313x.​2004.​02016.x
ship route line data and an SVM model, Kim et al. [104]
3. Sonoki T, Takahashi K, Sugita H, Hatamura M, Azuma Y, Sato
were able to forecast ship position, speed, and direction at T et al (2018) Glucose-free cis, cis-muconic acid production via
port locations. Parolas et al. [105] used SVM and ANN to new metabolic designs corresponding to the heterogeneity of
forecast the arrival time of container ships at the port of lignin. ACS Sustain Chem Eng 6(1):1256–1264
4. Kircher S, Schopfer P (2012) Photosynthetic sucrose acts as coty-
Rotterdam in other work that is concerned with on-time
ledon-derived long-distance signal to control root growth during
performance prediction. The findings showed that SVM early seedling development in Arabidopsis. Proc Natl Acad Sci
performed better than ANN. USA 109:11217–11221
Another notable area of the use of ML and DL in bio- 5. Durak H (2020) Hydrothermal liquefaction of Glycyrrhiza glabra
L. (Liquorice): effects of catalyst on variety compounds and
mass transportation is truck traffic prediction. In a work,
chromatographic characterization. Energy Sour Part A Recovery
Yuanchang Xie et al. [106] did research. GP and SVMs Util Environ Effects 42(20):2471–2484. https://​doi.​org/​10.​1080/​
were employed to anticipate the heavy truck traffic, since 15567​036.​2019.​16079​47
the heavy truck traffic produced by big seaports may have a 6. Durak H, Aysu T (2014) Effects of catalysts and solvents on
liquefaction of Onopordum heteracanthum for production of bio-
substantial impact on local and regional transportation net-
oils. Bioresour Technol 166:309–317. https://​doi.​org/​10.​1016/j.​
works. Another application area for ML and DL in transpor- biort​ech.​2014.​05.​051
tation may be regarded to be behavioral analysis and facility 7. Aysu T, Durak H (2016) Pyrolysis of giant mullein (Verbascum
design. To determine the most crucial aspects of transpor- Thapsus L.) in a fixed-bed reactor: Effects of pyrolysis parame-
ters on product yields and character. Energy Sour Part A Recover
tation cost, shipping distance, origin–destination, prod-
Util Environ Effects. 38(5):661–669. https://​doi.​org/​10.​1080/​
uct type, weight, and value, Mohri and Haghshenas [107] 15567​036.​2013.​819049
devised a DT. The most crucial inputs are hence shipping 8. Sun Y, Cheng J (2002) Hydrolysis of lignocellulosic materials
distance, perishability of the goods, and value of exports at for ethanol production: a review. Biores Technol 83:1–11
9. Molin U (2002) Importance of from hardwoods growing on
the destination. As a result, the following list includes the
southern pine sites cellulose/hemicellulose ratio for pulp
basic transportation applications of ML and DL that may strength. U.S. Dept. of Agric., For. Serv. GTR S0-24. Nordic
be utilized to carry biomass raw materials: operation and Pulp Paper Res J 17(1):14–19

13
 Journal of Material Cycles and Waste Management

10. Kartal F, Özveren U (2021) An improved machine learning impedance spectroscopy and machine learning. Sensors (Basel)
approach to estimate hemicellulose, cellulose, and lignin in bio- 20(4):1076. https://​doi.​org/​10.​3390/​s2004​1076.​PMID:​32079​
mass. Carbohydr Polymer Technol Appl 2:100148. https://​doi.​ 155;​PMCID:​PMC70​70823
org/​10.​1016/j.​carpta.​2021.​100148 26. Malek S, Miglietta F, Gobakken T, Næsset E, Gianelle D, Dal-
11. Kartal F, Özveren U (2022) Investigation of the chemical exergy ponte M (2019) Prediction of stem diameter and biomass at
of torrefied biomass from raw biomass by means of machine individual tree crown level with advanced machine learning
learning. Biomass Bioenergy 159:106383. https://​doi.​org/​10.​ techniques. iForest 12:323–329. https://​doi.​org/​10.​3832/​ifor2​
1016/j.​biomb​ioe.​2022.​106383 980-​012
12. Nag A, Gerritsen A, Doeppke C, Harman-Ware AE (2021) 27. Xue B (2012) Lidar and Machine Learning Estimation of Hard-
Machine learning-based classification of lignocellulosic bio- wood Forest Biomass in Mountainous and Bottomland Environ-
mass from pyrolysis-molecular beam mass spectrometry data. ments. Thesis, Northwest University, Bachelor of Science in
Int J Mol Sci 22(8):4107. https://​doi.​org/​10.​3390/​ijms2​20841​07. Management of Resource and Environment and Urban-Rural
(PMID: 33921121; PMCID: PMC8071563) Planning
13. Smuga-Kogut M, Kogut T, Markiewicz R, Słowik A (2021) Use 28. Martin-Luengo MA et al (2011) Renewable raw materials for
of machine learning methods for predicting amount of bioetha- advanced applications. World Congr Sustain Technol 2011:19–
nol obtained from lignocellulosic biomass with the use of ionic 22. https://​doi.​org/​10.​1109/​WCST1​9361.​2011.​61142​29
liquids for pretreatment. Energies 14(1):243. https://​doi.​org/​10.​ 29. Xue W, Hu X, Wei Z, Mei X, Chen X, Xu Y (2019) A fast and
3390/​en140​10243 easy method for predicting agricultural waste compost maturity
14. Kardani N, Hedayati Marzbali M, Shah K, Zhou A (2021) by image-based deep learning. Bioresour Technol 290:121761.
Machine learning prediction of the conversion of lignocel- https://​doi.​org/​10.​1016/j.​biort​ech.​2019.​121761
lulosic biomass during hydrothermal carbonization. Biofuels 30. Toklu E (2017) Biomass energy potential and utilization in
13:703–715 Turkey. Renew Energy 107:235–244. https://​doi.​org/​10.​1016/j.​
15. Karimi M, Khosravi M, Fathollahi R (2022) Determination of renene.​2017.​02.​008
the heat capacity of cellulosic biosamples employing diverse 31. Windrim L, Bryson M, McLean M, Randle J, Stone C (2019)
machine learning approaches. Wiley Online Publication Automated mapping of woody debris over harvested forest plan-
16. Fischer J, Lopes VS, Cardoso SL et al (2017) Machine learning tations using UAVs, high-resolution imagery, and machine learn-
techniques applied to lignocellulosic ethanol in simultaneous ing. Remote Sens 11(6):733. https://d​ oi.o​ rg/1​ 0.3​ 390/r​ s1106​ 0733
hydrolysis and fermentation. Braz J Chem Eng 34(1):53–63. 32. He T, Marco J, Soares R (2019) Machine learning models with
https://​doi.​org/​10.​1590/​0104-​6632.​20170​341s2​01504​75 quantitative wood anatomy data can discriminate between Swi-
17. Phromphithak S, Onsree T, Tippayawong N (2021) Machine etenia macrophylla and Swietenia mahagoni. J Forest 11:36
learning prediction of cellulose-rich materials from biomass 33. Chiang C, Barnes C, Angelov PP, Jiang RM (2020) Deep Learn-
pretreatment with ionic liquid solvents. Bioresour Technol ing based Automated Forest Health Diagnosis from Aerial
323:124642. https://​doi.​org/​10.​1016/j.​biort​ech.​2020.​124642 Images. IEEE Access 8:144064–144076
18. Sezer S, Kartal F, Özveren U (2022) Prediction of combustion 34. Silva JPM, da Silva MLM, de Mendonça AR, da Silva GF, de
reactivity for lignocellulosic fuels by means of machine learning. Barros Junior AA, da Silva EF, Aguiar MO, Santos JS, Rodri-
J Therm Anal Calorim 147:9793–9809. https://​doi.​org/​10.​1007/​ gues NMM (2021) Prognosis of forest production using machine
s10973-​022-​11208-8 learning techniques. Inf Process Agric. https://​doi.​org/​10.​1016/j.​
19. Löfgren J, Tarasov D, Koitto T, Rinke P, Balakshin M, Todorović inpa.​2021.​09.​004
M (2021) Lignin Biorefinery Optimization Through Machine 35. Yousafzai A, Manzoor W, Raza G, Mahmood T, Rehman F, Hadi
Learning. ChemRxiv. Cambridge Open Engage, Cambridge. This R, Shah S, Amin M, Akhtar A, Bashir S, Habiba U, Hussain M
content is a preprint and has not been peer-reviewed (2021) Forest yield prediction under different climate change
20. Kibleur P, Aelterman J, Boone MN, Van den Bulcke J, Van Acker scenarios using data intelligent models in Pakistan. Braz J Biol
J (2022) Deep learning segmentation of wood fiber bundles in 84:e253106. https://d​ oi.o​ rg/1​ 0.1​ 590/1​ 519-6​ 984.2​ 53106. (PMID:
fiberboards. Compos Sci Technol 221:109287. https://​doi.​org/​ 34730700)
10.​1016/j.​comps​citech.​2022.​109287 36. Rana P, Miller D (2019) Machine learning to analyze the social-
21. Pierdicca R, Balestra M, Micheletti G, Felicetti A, Toscano G ecological impacts of natural resource policy: insights from com-
(2022) Semi-automatic detection and segmentation of wooden munity forest management in the Indian Himalaya. Environ Res
pellet size exploiting a deep learning approach. Renewable Lett 14:024008
Energy 197:406–416. https://​doi.​org/​10.​1016/j.​renene.​2022.​07.​ 37. Tian H, Zhu J, He X, Chen X, Jian Z, Li C, Ou Q, Li Q, Huang
109 G, Liu C, Xiao W (2022) Using machine learning algorithms to
22. Plankenbuhler T et al (2020) “Image-Based Model for Assess- estimate stand volume growth of Larix and Quercus forests based
ment of Wood Chip Quality and Mixture Ratios.” Processes, on national-scale Forest Inventory data in China. Forest Ecosyst.
8(6). Gale Academic OneFile. http://​link.​gale.​com/​apps/​doc/​ 9:100037. https://​doi.​org/​10.​1016/j.​fecs.​2022.​100037
A6340​50532/​AONE?u=​anon~36401​5e&​sid=​sitem​ap&​xid=​ 38. Zhang F, Tian X, Zhang H, Jiang M (2022) Estimation of above-
cded6​9b6. Accessed 13 Oct 2022 ground carbon density of forests using deep learning and mul-
23. Pitak L, Laloon K, Wongpichet S, Sirisomboon P, Posom J tisource remote sensing. Remote Sens 14(13):3022. https://​doi.​
(2021) Machine learning-based prediction of selected param- org/​10.​3390/​rs141​33022
eters of commercial biomass pellets using line scan near infrared- 39. Hart E, Sim K, Kamimura K, Meredieu C, Guyon D, Gardiner B
hyperspectral image. Processes 9(2):316. https://d​ oi.o​ rg/1​ 0.3​ 390/​ (2019) Use of machine learning techniques to model wind dam-
pr902​0316 age to forests. Agric For Meteorol 265:16–29. https://​doi.​org/​10.​
24. Oehmcke S, Li L, Revenga J, Nord-Larsen T, Trepekli K (2021) 1016/j.​agrfo​rmet.​2018.​10.​022 (ISSN 0168-1923)
Deep Learning Based 3D Point Cloud Regression for Estimating 40. Saha S, Bhattacharjee S, Shit PK, Sengupta N, Bera B (2022)
Forest Biomass. J Computer Vision Pattern Recognit. Cornell Deforestation probability assessment using integrated machine
University Publication learning algorithms of Eastern Himalayan foothills (India).
25. Tiitta M, Tiitta V, Heikkinen J, Lappalainen R, Tomppo Resour Conserv Recycl Adv. 14:200077. https://​doi.​org/​10.​
L (2020) Classification of wood chips using electrical 1016/j.​rcradv.​2022.​200077

13
Journal of Material Cycles and Waste Management

41. Wan L, Li S, Chen Y, He Z, Shi Y (2022) Application of deep 57. Cubillos M (2020) Multi-site household waste generation fore-
learning in land use classification for soil erosion using remote casting using a deep learning approach. Waste Manag 115:8–14.
sensing. Front Earth Sci 10:849531. https://d​ oi.o​ rg/1​ 0.3​ 389/f​ eart.​ https://​doi.​org/​10.​1016/j.​wasman.​2020.​06.​046
2022.​84953 58. Mao WL, Chen WC, Wang CT, Lin YH (2021) Recycling waste
42. Munis RA, Almeida RO, Camargo DA, da Silva RBG, classification using optimized convolutional neural network.
Wojciechowski J, Simões D (2022) Machine learning methods to Resour Conserv Recycl 164:105132. https://​doi.​org/​10.​1016/j.​
estimate productivity of harvesters: mechanized timber harvest- resco​nrec.​2020.​105132
ing in Brazil. Forests 13(7):1068. https://​doi.​org/​10.​3390/​f1307​ 59. Ping P, Xu G, Kumala E, Gao J (2020) Smart Street litter detec-
1068 tion and classification based on faster R-CNN and edge comput-
43. Lee J, Im J, Kim K, Quackenbush LJ (2018) Machine learning ing. Int J Softw Eng Knowl Eng 30(4):537–553. https://​doi.​org/​
approaches for estimating forest stand height using plot-based 10.​1142/​S0218​19402​04000​45
observations and airborne LiDAR data. Forests 9(5):268. https://​ 60. Kubanek M, Bobulski J (2019) Waste classification system using
doi.​org/​10.​3390/​f9050​268 image processing and convolutional neural networks. Int. Work.
44. Zhang H, Yin A, Yang X, Fan M, Shao S, Wu J, Wu P, Zhang M, Artif. Neural Networks. Springer, Cham, pp 350–361. https://d​ oi.​
Gao C (2021) Use of machine-learning and receptor models fo org/​10.​1007/​978-3-​030-​20518-8_​30
prediction and source apportionment of heavy metals in coastal 61. Hussain A et al (2020) Waste management and prediction of air
reclaimed soils. Ecol Indic 122:107233. https://d​ oi.o​ rg/1​ 0.1​ 016/j.​ pollutants using iot and machine learning approach. Energies
ecoli​nd.​2020.​107233 13(15):3930–3951
45. Furuya DEG, Aguiar JAF, Estrabis NV, Pinheiro MMF, Furuya 62. Liang S, Gu Y (2021) A deep convolutional neural network to
MTG, Pereira DR, Gonçalves WN, Liesenberg V, Li J, Marcato simultaneously localize and recognize waste types in images.
Junior J, Prado Osco L, Ramos APM (2020) A machine learning Waste Manag 126:247–257. https://​doi.​org/​10.​1016/j.​wasman.​
approach for mapping forest vegetation in Riparian Zones in an 2021.​03.​017
Atlantic biome environment using sentinel-2 imagery. Remote 63. Panwar H et al (2021) AquaVision: automating the detection of
Sens 12(24):4086. https://​doi.​org/​10.​3390/​rs122​44086 waste in water bodies using deep transfer learning. Case Stud
46. Chen X, Avtar R, Umarhadi DA, Louw AS, Shrivastava S, Yunus Chem Environ Eng. https://d​ oi.o​ rg/1​ 0.1​ 016/j.c​ scee.2​ 020.1​ 00026
AP, Khedher KM, Takemi T, Shibata H (2022) Post-typhoon 64. Sterkens W, Diaz-Romero D, Goedemé T, Dewulf W, Peeters JR
forest damage estimation using multiple vegetation indices and (2021) Detection and recognition of batteries on X-Ray images
machine learning models. Weather Clim Extremes 38:100494 of waste electrical and electronic equipment using deep learning.
47. Dantas D, Terra MD, Schorr LP, Calegário N (2021) Machine Resour Conserv Recycl 168:105246. https://​doi.​org/​10.​1016/j.​
learning for carbon stock prediction in a tropical forest in South- resco​nrec.​2020
eastern Brazil. Bosque, Valdivia 65. Birgen C, Magnanelli E, Carlsson P, Skreiberg Ø, Mosby
48. Han Z, Sun J, Lv A, Sung Y, Shi H, Hu X, Xing K (2017) Isola- J, Becidan M (2021) Machine learning based modelling for
tion, identification and characterization of Shewanella algae from lower heating value prediction of municipal solid waste. Fuel
reared tongue sole, Cynoglossus semilaevis Günther. Aquaculture 283:118906. https://​doi.​org/​10.​1016/j.​fuel.​2020.​118906
468:356–362 66. Bijos JCBF, Zanta VM, Morató J, Queiroz LM, Oliveira-
49. Khan R, Kumar S, Srivastava AK, Dhingra N, Gupta M, Bhati N, Esquerre KPSR (2022) Improving circularity in municipal solid
Kumari P (2021) Machine learning and IoT-based waste manage- waste management through machine learning in Latin America
ment model. Comput Intell Neurosci 2021:5942574. https://​doi.​ and the Caribbean. Sustain Chem Pharm 28:100740. https://​doi.​
org/​10.​1155/​2021/​59425​74.​PMID:​34484​322;​PMCID:​PMC84​ org/​10.​1016/j.​scp.​2022.​100740
16408 67. Kabugo JC, Jämsä-Jounela S-L, Schiemann R, Binder C (2020)
50. Sevcik M, Skvaril J, Tomas Aparicio E (2019) Applications of Industry 4.0 based process data analytics platform: a waste-
hyperspectral imaging and machine learning methods for real- to-energy plant case study. Int J Electr Power Energy Syst
time classification of waste stream components. Presented at the 115:105508. https://​doi.​org/​10.​1016/j.​ijepes.​2019.​105508
19th International Conference on Near Infrared Spectroscopy 68. Gustavsson J, Cederberg C, Sonesson U, Van Otterdijk R, Mey-
- NIR 2019, Gold Coast, Australia, 15–20 September 2019 beck A (2011) Global food losses and food waste. FAO, Rome
51. Adawiyah AR, Ibrahim NN (2022) A hybrid P-graph and WEKA 69. Basher SA, Raboy D, Kaitibie S, Hossain I (2013) Understanding
approach in decision-making: waste conversion technologies challenges to food security in dry arab micro-states: evidence
selection. J Appl Sci Eng 26(2):261–267 from qatari micro-data. Social Science Research Network, Roch-
52. Huang GL, He J, Xu Z, Huang G (2020) A combination model ester, NY
based on transfer learning for waste classification. Concurr Com- 70. Irani Z, Sharif AM (2016) Sustainable food security futures: per-
put 32(19):1–12. https://​doi.​org/​10.​1002/​cpe.​5751 spectives on food waste and information across the food supply
53. Vo AH, Hoang Son L, Vo MT, Le T (2019) A novel framework chain. J Enterp Inf Manag 29:171–178. https://​doi.​org/​10.​1108/​
for trash classification using deep transfer learning. IEEE Access JEIM-​12-​2015-​0117
7:178631–178639. https://d​ oi.o​ rg/1​ 0.1​ 109/A
​ CCESS.2​ 019.2​ 9590​ 71. Barad M, Braha D (1996) Control limits for multi-stage manufac-
33 turing processes with binomial yield (single and multiple produc-
54. Kang Z, Yang J, Li G, Zhang Z (2020) An automatic garbage clas- tion runs). J Oper Res Soc 47:98–112. https://​doi.​org/​10.​1057/​
sification system based on deep learning. IEEE Access 8:140019– jors.​1996.9
140029. https://​doi.​org/​10.​1109/​ACCESS.​2020.​30104​96 72. Garre A, Ruiz MC, Hontoria E (2020) Application of machine
55. Bai J, Lian S, Liu Z, Wang K, Liu D (2019) Deep learning based learning to support production planning of a food industry in the
robot for automatically picking up garbage on the grass. IEEE context of waste generation under uncertainty. Op Res Perspect.
Trans Consum Electron 64(3):382–389 7:100147. https://​doi.​org/​10.​1016/j.​orp.​2020.​100147
56. Kumar S, Yadav D, Gupta H, Verma OP, Ansari IA, Ahn CW 73. Czarnecki S, Hadzima-Nyarko M, Chajec A et al (2022)
(2021) A novel yolov3 algorithm-based deep learning approach Design of a machine learning model for the precise manufac-
for waste segregation: towards smart waste management. Elec- turing of green cementitious composites modified with waste
tron 10(1):1–20. https://​doi.​org/​10.​3390/​elect​ronic​s1001​0014 granite powder. Sci Rep 12:13242. https://​doi.​org/​10.​1038/​
s41598-​022-​17670-6

13
 Journal of Material Cycles and Waste Management

74. Bhandari S (2020) Automative Waste Sorting in Industrial Envi- 90. Ahmad F, Younis MS, Zahid RU, Shahid LA (2020) Machine
ronment via Machine Learning Approaches., Information Tech- learning based grain moisture estimation for real-time monitor-
nology and Communication Sciences TAU, Master's Thesis ing of high-temperature paddy drying silo. In: 2020 IEEE 23rd
75. Lee HY, Choi HN, Jung YJ, Uhm HS, Kang BK (2003) “Sewage International Multitopic Conference (INMIC). Bahawalpur, Paki-
sludge treatment by arc discharge,” Digest of Technical Papers. stan, pp 1–6. https://​doi.​org/​10.​1109/​INMIC​50486.​2020.​93180​
PPC-2003. In: 14th IEEE International Pulsed Power Conference 71
(IEEE Cat. No.03CH37472), vol 2. pp 1247–1249. https://​doi.​ 91. Sivasothy P, Andres M, Corbin G (2018) Machine learning based
org/​10.​1109/​PPC.​2003.​12780​39 filling level estimation for bulk solid silos. J Acoust Soc Am
76. Zambrano J, Samuelsson O, Carlsson B (2019) Machine learn- 144:1685. https://​doi.​org/​10.​1121/1.​50674​92
ing techniques for monitoring the sludge profile in a second- 92. Serva L, Magrin L, Marchesini G, Andrighetto I (2022) Short
ary settler tank. Appl Water Sci 9:146. https://​doi.​org/​10.​1007/​ communication: prognostic values of a multiparametric risk
s13201-​019-​1018-5 score in maize silage undergoing different ensiling conditions.
77. Boniecki P, Dach J, Pilarski K, Piekarska-Boniecka H (2012) Agronomy 12(4):774. https://​d oi.​o rg/​1 0.​3 390/​a gron​o my12​
Artificial neural networks for modeling ammonia emissions 040774
released from sewage sludge composting. Atmos Environ 57:49– 93. Farea S, Kanaan M (2019) Support vector machine classifier of
54. https://​doi.​org/​10.​1016/j.​atmos​env.​2012.​04.​036 grain stored in silo-bags by using wireless network of tempera-
78. Kujawa S, Mazurkiewicz J, Czekała W (2020) Using convolu- ture and moisture sensors. In: International congress on human-
tional neural networks to classify the maturity of compost based computer interaction, optimization and robotic applications, vol
on sewage sludge and rapeseed straw. J Clean Prod 258:120814. 4, no 5, pp 11–14
https://​doi.​org/​10.​1016/j.​jclep​ro.​2020.​120814 94. Kowalczyk-Juśko A, Pochwatka P, Zaborowicz M, Czekała W,
79. Safder U, Loy-Benitez J, Nguyen HT, Yoo C (2022) A hybrid Mazurkiewicz J, Mazur A, Janczak D, Marczuk A, Dach J (2020)
extreme learning machine and deep belief network framework Energy value estimation of silages for substrate in biogas plants
for sludge bulking monitoring in a dynamic wastewater treat- using an artificial neural network. Energy 202:117729. https://​
ment process. J Water Process Eng. 46:102580. https://​doi.​org/​ doi.​org/​10.​1016/j.​energy.​2020.​117729
10.​1016/j.​jwpe.​2022.​102580 95. Duysak H, Yigit E (2020) Machine learning based quantity
80. Araujo M, Rivas T, Giraldez E, Taboada J (2011) Use of machine measurement method for grain silos. Measurement 152:107279.
learning techniques to analyse the risk associated with mine https://​doi.​org/​10.​1016/j.​measu​rement.​2019.​107279
sludge deposits. Math Comput Model 54(7–8):1823–1828. 96. Lutz É, Coradi PC (2022) Applications of new technologies for
https://​doi.​org/​10.​1016/j.​mcm.​2010.​11.​066 monitoring and predicting grains quality stored: sensors, internet
81. Goldstein H (2002) Waste not, pollute not [animal waste recy- of things, and artificial intelligence. Measurement 188:110609.
cling for power production]. IEEE Spectr 39(1):72–77. https://​ https://​doi.​org/​10.​1016/j.​measu​rement.​2021.​110609
doi.​org/​10.​1109/6.​975026 97. Zahraee SM, Golroudbary SR, Shiwakoti N, Stasinopoulos P
82. I Pence, K Kumaş, MC Siseci et al (2022) Detailed analysis of (2022) Palm oil biomass global supply chain: environmental
animal manure-based CO2 emissions, coal, electricity, thermal emissions vs. technology development of maritime transpor-
energy, and CH4 emissions and using machine learning as a tation. Procedia CIRP 105:817–822. https://​doi.​org/​10.​1016/j.​
forecasting method: a study from Turkey. PREPRINT (Version procir.​2022.​02.​135
1) available at Research Square. https://​doi.​org/​10.​21203/​rs.3.​ 98. Milenković M, Milosavljevic N, Bojović N, Val S (2021)
rs-​16482​18/​v1 Container flow forecasting through neural networks based on
83. Matthews SG, Miller AL, PlÖtz T, Kyriazakis I (2017) Auto- metaheuristics. Oper Res 21(2):965–997
mated tracking to measure behavioural changes in pigs for health 99. Niu M, Hu Y, Sun S, Liu Y (2018) A novel hybrid decomposi-
and welfare monitoring. Sci Rep 7:17582 tion-ensemble model based on VMD and HGWO for container
84. Alonso J, Villa A, Bahamonde A (2015) Improved estimation throughput forecasting. Appl Math Model 57:163–178. https://​
of bovine weight trajectories using support vector machine clas- doi.​org/​10.​1016/j.​apm.​2018.​01.​014
sification. Comput Electron Agric 110:36–41 100. Loaiza MF, Solano RP, Simancas R, Ojito VH (2017) Modeling
85. Xue B (2015) Lidar and Machine Learning Estimation of Hard- demand for air cargo in the Colombian context. Int Conf Adv
wood Forest Biomass in Mountainous and Bottomland Environ- Mater Sci Civil Eng 70:132–137
ments. Theses and Dissertations. 1274. http://s​ chola​ rwork​ s.u​ ark.​ 101. Du Y, Meng Q, Wang S, Kuang H (2019) Two-phase optimal
edu/​etd/​1274 solutions for ship speed and trim optimization over a voy-
86. Septiyadi E, Firmansyah E, Danang Wijaya F (2017) Prelimi- age using voyage report data. Transp Res Part B Methodol
nary design of biomass fuel storage on a thermoelectric based 122:88–114
biomass-electric energy conversion. In: 2017 7th International 102. Gao Y, Chang D, Chen CH, Fang T (2018) Deep learning with
Annual Engineering Seminar (InAES). Yogyakarta, Indonesia, long short-term memory recurrent neural network for daily con-
pp 1–5. https://​doi.​org/​10.​1109/​INAES.​2017.​80685​72 tainer volumes of storage yard predictions in port. 2018 Interna-
87. Farahani RZ, Asgari N, Hojabri H, Jaafari AA (2009) Optimizing tional Conference on Cyberworlds. IEEE, pp 427–430
wheat storage and transportation system using a mixed integer 103. Shyur HJ, Luxhoj JT, Williams TP (1996) Using neural networks
programming model and genetic algorithm: a case study. IEEE to predict component inspection requirements for aging aircraft.
Int Conf Ind Eng Eng Manag 2009:2109–2113. https://​doi.​org/​ Comput Ind Eng 30(2):257–267
10.​1109/​IEEM.​2009.​53731​52 104. Kim JS (2017) Vessel target prediction method and dead reck-
88. Wang G, Li Q, You J (2008) On wheat storage period estimation oning position based on SVR seaway model. Int J Fuzzy Logic
using Markov_chain Analysis. In: 2008 6th IEEE International Intell Syst 17(4):279–288
Conference on Industrial Informatics. Daejeon, pp 882–887. 105. Parolas I, Tavasszy L, Kourounioti I, van Duin R, Cities K (2017)
https://​doi.​org/​10.​1109/​INDIN.​2008.​46182​25 Prediction of vessel’s estimated time of arrival (ETA) using
89. Goettsch D, Castillo-Villar KK, Aranguren M (2020) Machine- machine learning—a port of Rotterdam case study. 96th Annual
learning methods to select potential depot locations for the sup- Meeting of the Transportation Research. pp 8–12
ply chain of biomass co-firing. Energies 13(24):6554. https://d​ oi.​
org/​10.​3390/​en132​46554

13
Journal of Material Cycles and Waste Management

106. Xie Y, Huynh N (2010) Kernel-based machine learning models Springer Nature or its licensor (e.g. a society or other partner) holds
for predicting daily truck volume at seaport terminals. J Transp exclusive rights to this article under a publishing agreement with the
Eng 136(12):1145–1152 author(s) or other rightsholder(s); author self-archiving of the accepted
107. Mohri S, Haghshenas H (2017) Modeling the container selection manuscript version of this article is solely governed by the terms of
for freight transportation: case study of Iran. Int Jof Transp Eng such publishing agreement and applicable law.
5(1):31–44

Publisher's Note Springer Nature remains neutral with regard to

jurisdictional claims in published maps and institutional affiliations.

13