Received 26 December 2022, accepted 25 January 2023, date of publication 3 February 2023, date of current version 10 February 2023.

Digital Object Identifier 10.1109/ACCESS.2023.3242547

Garbage Content Estimation Using Internet of

Things and Machine Learning
EUNICE LIKOTIKO , YUKI MATSUDA , (Member, IEEE),
AND KEIICHI YASUMOTO , (Member, IEEE)
Nara Institute of Science and Technology, Nara 630-0192, Japan
Corresponding author: Eunice Likotiko ([email protected])
This work was supported by the Japan Science and Technology Agency (JST), Precursory Research for Embryonic Science and
Technology (PRESTO), under Grant JPMJPR2039.
This work involved human subjects or animals in its research. Approval of all ethical and experimental procedures and protocols was
granted by the Ethics Review Board at the Nara Institute of Science and Technology under Approval No. 2019-I-9-2.

ABSTRACT Much garbage is produced daily in homes due to living activities, including cooking and eating.
The garbage must be adequately managed for human well-being and environmental protection. Although
the existing IoT-based smart garbage systems have gained high garbage classification accuracy, they still
have a problem that they provide a small number of garbage categories, not enough for reasonable practices
of household garbage separation. This study presents a new smart garbage bin system, SGBS, embedded
with multiple sensors to solve the problem. We deployed temperature, humidity, and gas sensors to know the
condition and identify the garbage content disposed of. Then, we introduce a new garbage content estimation
method by training a machine learning model using daily collected fuse sensor readings combined with
detailed household garbage contents annotations to perform garbage classification tasks. For evaluation,
we deployed the designed SGBS in five households over one month. As a result, we confirmed that the
leave-one-house cross-validation results showed an accuracy of 91% in 5 kitchen waste contents, also, 89%
in 5 paper/softbox contents, and 85% in the 8 garbage categories for the classification tasks.

INDEX TERMS IoT-based smart garbage system, garbage content estimation, machine learning algorithms.

I. INTRODUCTION it would improve garbage management services through

Much garbage is produced daily in homes due to living proper garbage separation practices for the well-being of
activities, including cooking and eating. Therefore, garbage people and the environment.
must be adequately managed for human well-being and It is reported that the world generates 2.01 billion tonnes
environmental protection. In the standard municipal garbage of municipal solid waste annually, with at least 33% of
management system, households are responsible for sorting that not managed environmentally safely [1]. In fact, daily
and managing garbage produced in their home. However, waste generated per person ranges widely, from 0.11 to
it is hard to depend solely on public awareness to provide 4.54 kilograms [2]. Furthermore, only 17% of electronic
the correct garbage management at the source. Therefore, garbage is collected and recycled [3]. Moreover, 32% of plas-
an automation tool that can reflect the home’s daily life and tic packages still need to be managed, which leads to severe
understand households’ routine behaviour of garbage dis- implications for ecological balance and human well-being.
posal would be necessary to influence behaviour change on But, again, garbage separation by the person who disposes of
garbage disposal and increase home monitoring for the case garbage has been widely accepted as ethical behaviour and
of elderly anomaly detection and healthy living. Furthermore, best practice for reducing, reusing, and recycling [4]. Several
existing IoT-based smart garbage systems and the classifi-
The associate editor coordinating the review of this manuscript and cation methods using computer vision and artificial intelli-
approving it for publication was Nikhil Padhi . gence have been developed to improve household garbage

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
13000 VOLUME 11, 2023
E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

management [5], [6], [7], [8]. However, the existing sys- of in households, built with data-efficiency machines
tems have the following problems: first, they can not learn learning classifiers with satisfactory relative accuracy.
the amount of garbage disposed of each time; second, they The remainder of this paper is as follows: Section II pro-
provide a small number of garbage categories, not enough vides an overview of related work from the recent work on
for reasonable practices of household garbage separation; garbage classification using the image and Deep learning
and third, they can not understand the routine behaviour of models also Municipal garbage separation rules. Section III
garbage disposal by households. describes the materials and tools used in the study, including
In our previous study [9], we addressed the first problem systems design and development details. Section IV presents
by proposing a smart garbage bin system with ToF and weight the experiment, data collection and pre-processing data pro-
sensors and the ARIMA model based garbage growth predic- cedures. Section V introduces the garbage content estimation
tion method. In this paper, we focus on solving the second model and the step by steps process of building the model
and the third problems, we propose a newly designed and using a machine learning algorithms. Finally, Section VI
developed smart garbage bin system (SGBS) embedded with discusses results from the classification tasks and compares
multiple sensors to identify the garbage contents disposed of. our approach with literature works, whereas, in Section VII,
The SGBS architecture comprised two subsystems. we conclude our paper.
The first subsystem is the smart garbage bin (SGB),
embedded with DHT22 (temperature and humidity) and
II. RELATED WORK
MQ135 gas sensors to know the conditions and identify
the disposed garbage content since garbage contents have This section gives an overview of related work from two
different shapes and moisture. Therefore, the type of garbage different perspectives. First, we provide an overview of the
content affects the humidity and air quality found in the smart separation and disposal of garbage with an emphasis on
bin. Also, the SGB is embedded with ToF (time of flight) and municipals in Japan, where this study was conducted. Sec-
load cell sensors to detect the new garbage content disposed ondly, we discuss recent work on garbage classification from
of each time. Then, data are updated and stored in the cloud images using deep learning to recall existing approaches to
via a Wi-Fi gateway. assess it. Thirdly we briefly discuss our preliminary study.
The second subsystem is a garbage annotation mobile
application (GAA). The GAA interface consists of 8 garbage A. SEPARATION AND DISPOSAL OF GARBAGE IN JAPAN
categories and 25 garbage content identities, providing an Garbage separation has been a major challenge across devel-
easy way for household users to annotate garbage content oping countries than in developed countries where there are
they dispose of daily using a handy smartphone. various collection systems for house-separated garbage, such
We conducted experiments where the SGBS was deployed as in Sweden and Germany [10], China [11], and Japan [12].
in five houses of heterogeneous characteristics to examine the While in other developed countries, garbage separation is
impact. As a result, the household user daily uses the installed often classified into three categories: recyclable, household,
smart garbage bin system and annotates their garbage con- and vegetation garbage. In Japan, the garbage separation and
tents, which they dispose of in smart garbage bins. Therefore, disposal system is different and complex. The rules for the
information about identified garbage and produced amounts separation and disposal of garbage depend on the particular
were continuously monitored and collected in the garbage log local municipality, whereby each city in Japan provides a
for each household. To perform garbage classification tasks, well-documented pamphlet explaining the garbage disposal
we introduce a new garbage content estimation method by rules. In general, garbage is divided into four categories:
training a machine learning model using daily collected fuse Burnable garbage (Kitchen waste, paper scraps, clothing,
sensor readings combined with detailed household garbage etc.), non-burnable garbage (Metal, glass, ceramics and pot-
contents annotations. As a result, we confirmed that the leave- tery, etc.), recyclable (Plastic bottles, container jars, cans,
one-house-out cross-validation results showed an accuracy of newspapers, etc.), and oversized (Large furniture, etc.) [12].
91% in 5 kitchen waste contents, also, 89% in 5 paper/softbox Therefore, each municipality uses such a general garbage
contents, and 85% in the 8 garbage categories for the clas- division to classify garbage for their residents. Table 1 pro-
sification tasks. In summary, the contributions of this work vides an overview of the division of burnable garbage con-
are: tent in four cities in Japan: Kashihara [13], Ikoma [14],
1) Identification of garbage content and understanding Nara [15] and Kyoto [16]. Apart from garbage descriptions
household garbage disposal behaviour for influencing from the municipal pamphlets, residents use designated plas-
family’s behaviour change in the garbage disposal and tic garbage bags of up to 45 litres to dispose of garbage.
increase home monitoring. Moreover, garbage collection for each category of garbage is
2) The provision of more satisfactory garbage content set by the municipal for instance, Mondays and Thursdays
categories for the reasonable practice of separating in Ikoma city [14] are used for the collection of burnable
garbage in the household. garbage only. The above facts show that families in Japan play
3) Providing and discussing a new garbage content esti- a hand role in their municipal rules for garbage separation
mation model based on daily garbage contents disposed and disposal systems. However, the failure of households

VOLUME 11, 2023 13001

 E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

TABLE 1. Overview of burnable garbage separation in Japan. Besides, there is still a shortage of publicly available garbage
image datasets and an information gap in their experimental
procedures.
Furthermore, Wang et al. [7] revealed garbage sorting and
classification at the source, the beginning of garbage collec-
tion while utilizing the combined method of IoT and CNN.
The study used experimental data available in the Trash-
net [23] dataset, merged with other datasets thus, resulted
in nine categories of garbage (Kitchen waste, other waste,
hazardous waste, plastic, glass, paper or cardboard, metal,
fabric and other recyclable waste). In addition, the study
developed an intelligent bin embedded with ultrasonic sen-
sors, MQ9, and MQ135 gas sensors to monitor the garbage’s
running state in the bin. Finally, the CNN model was
deployed in mobile phones and cloud computing servers
for garbage classification. The system required citizens to
take pictures of garbage using their mobile phones and send
them to a cloud server to run the deep-learning algorithm
to recognize categories. Despite the high-performance accu-
racies of 92.44% and 92.00% achieved by Xception and
to sort the garbage renders the whole system useless [7]. MobileNetV3 models on classifying nine types of garbages,
Therefore, automation tools are necessary to monitor daily the author presented more generalizable garbage categories
family garbage disposal and improve garbage separation and that need to be improved for proper household garbage
management. separation.
Besides, a distributed architecture for smart recycling using
B. GARBAGE CLASSIFICATION FROM IMAGES WITH DEEP machine learning was realized by Ziouzios et al. [6] as a solu-
LEARNING MODELS tion for garbage classification in collection facilities to solve
A possible solution to overcome the existing challenges in the problem of non-segregated garbage, which exists more
household garbage separation and management is to adopt in developing and developed countries. The Trashnet [23]
sustainable automation tools to improve garbage separation. dataset was used for training the models by utilizing computa-
Presently, several works have been devoted to the automa- tion offloading to the cloud. The CNN architecture classified
tion and detection of garbage from images, which has now the garbage materials into five categories: paper, glass, plas-
become a popular choice to replace manual garbage sep- tic, metal, carton, and trash. Similarly, Sami et al. [24] used
aration while taking advantage of the rapid advances in the Trashnet [23] dataset to automate the garbage classifica-
computer vision and artificial intelligence. Various stan- tion problem into six classes: glass, paper, metal, cardboard,
dard CNN architectures have been recently proposed to and trash using a Support Vector Machine, Random Forest,
perform image classification tasks with high accuracies, Decision tree, and CNN to find the optimal algorithm that best
such as VGGNet [17], AlexNet [18], ResNet [19] and fits garbage classification solution. However, the available
DenseNet [20]. public garbage image datasets need more classes of garbage
Nnamoko et al. [5] investigated the problem of manual categories for proper garbage classification. Therefore, the
household garbage separation into two categories, namely, garbage categories presented in both studies [6], [24] are not
organic and recyclable. Experiments presented in this paper practical for household garbage separation and for improving
were conducted with Sekar’s waste classification image the garbage management systems.
dataset available in the Kaggle library [21]. Later, a bespoke Despite the high accuracies achieved by the existing solu-
5-layer CNN architecture was used to perform image clas- tions on garbage classification through the automation and
sification tasks. In this work, the training was conducted detection of garbage from images by the deep learning mod-
on two datasets, smaller model (80 × 45 pixels) and a els, they still have problems: (Problem 1) They can not learn
larger model (225 × 264 pixels), for performance com- the amount of garbage disposed of each time; (Problem 2)
parison, thus obtaining similar cross-validation accuracy of They provide a small number of garbage categories, not
79%. Likewise, Mookkaiah et al. [22] proposed a model to enough for reasonable practices of household garbage sep-
identify and classify two types of garbage, biodegradable aration; (Problem 3) They can not understand households’
and non-biodegradable. First, the images were collected in routine behaviour of garbage disposal. Therefore, to the best
the respective garbage bin by Raspberry Pi Camera Mod- of our knowledge, an automation tool that can learn and
ule v2. Then garbage classification task was done by CNN identify the daily garbage content disposed of in homes and
architecture. However, separating garbage into two categories perform classification tasks, as investigated throughout this
is insufficient for logical household garbage separation. work, has yet to be considered.

13002 VOLUME 11, 2023

E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

C. PRELIMINARY STUDY TABLE 2. Sensor used in development of smart garbage bin.

To solve Problem 1, we conducted a preliminary study to

learn the amount of garbage disposed of each time and pre-
dict growth behaviour at a single house [9]. In this study,
we designed and developed the initial smart garbage bin
prototype embedded with ToF (time of flight) and load cell
sensors to track the amount of garbage during disposal. Using
a Wi-Fi gateway, data were sent to a cloud platform. For
evaluation, we deployed the smart garbage bin in a stu- detect the timestamp of newly disposed of garbage content
dent laboratory over one month. An autoregressive integrated during garbage disposal. On the other hand, the smart garbage
moving average (ARIMA) model was applied, providing an bin (SGB) is embedded with temperature, humidity, and gas
average mean absolute error (MAE) of 5.17 cm and a standard sensors to identify and distinguish disposed of garbage con-
deviation (SD) of 0.33 cm, thus was considered satisfactory tents. Secondly, SGBS architecture comprises the garbage
accuracy for the garbage growth prediction. Therefore, our annotation mobile application (GAA) with a smooth interface
prediction model was suitable for predicting future garbage that allows users to annotate their daily disposal of garbage
growth behaviour, enhancing flexibility in the garbage col- content during garbage disposal. The two subsystems (SGB
lection schedule and the frequency of changing garbage bags and GAA) later create a daily garbage log data for each house.
in the smart bin. Moreover, the designed architecture comprises the analysis
However, Problem 2 and Problem 3 in Section II-B remain part that uses machine learning algorithms to classify garbage
open. Therefore, in this paper, we try to address these contents found in the house logs. The outcome of the analysis
problems. produces a garbage content estimator for each home which
helps identify and classify garbage content at the source.
III. MATERIALS AND TOOLS
This section presents the details of the system requirements C. SMART GARBAGE BIN
necessary for designing and developing a smart garbage bin Fig. 2 shows the overview of a designed and developed
system (SGBS), tools and the procedure for selecting impor- smart garbage bin system (SGBS). Considering the signif-
tant garbage categories for developing garbage annotation icant roles of the proposed SGBS architecture described in
application design. Section III-B, a set of lightweight, low-cost, high-precision
IoT sensors were chosen and embedded in the smart garbage
A. SYSTEM REQUIREMENTS bin (SGB). The selected devices have different hardware
configurations and purposes. In our SGB prototype, we used
In this subsection, we describe the system requirements for
a DHT22 (temperature and humidity) and MQ135 gas sen-
the proposed system. Based on the discussions in Section I
sors to monitor the moisture and air quality of the disposed
and Section II, we find the following two requirements for a
garbage content in the smart garbage bin. Furthermore,
smart garbage bin system:
we used a ToF (time of flight) and HX711-load cell to
1) The smart garbage bin system should automatically track the garbage filling level and weight at each time of
collect sensor data without any additional activities by disposal. Using a Wi-Fi gateway, the smart garbage bin sys-
users. tem is always connected to the internet, uploads all sensor
2) The smart garbage bin system should estimate detailed data to the cloud, and stores them. In addition, the Secure
garbage categories and garbage content identities cor- Digital non-volatile flash memory card format (SD), con-
responding to each disposal behaviour. nected to an I2C real-time clock with 32.768 kHz frequency
To address requirement (1), we designed and developed a (DS3231 RTC) module data are also collected and stored in
smart garbage bin system which is always connected to the the SD-created file in one-minute intervals daily. On the other
internet, uploads all sensor data to the cloud to store them. hand, the SGB comprises the 2 × 16 character LCD Module
To address requirement (2), we built a new machine learning with a blue backlight, which uses an I2C interface to com-
model for estimating garbage categories and garbage content municate with the host Arduino Mega 2560 microcontroller
identities with high accuracy. Rev3. Therefore, the LCD module displays the garbage’s
current filling level and temperature data of the smart bin. The
B. ARCHITECTURE DESIGN proposed smart garbage bin prototype allows easy tracking of
Fig. 1 demonstrates a designed and developed SGBS archi- garbage amount information at the source. Table 2 provides
tecture to revolutionize the existing household garbage man- the purpose of the chosen sensors used to develop the smart
agement system by tracking daily household garbage disposal garbage bin.
information and identifying the type of garbage contents
disposed of at the source. The smart garbage bin system D. GARBAGE ANNOTATION APPLICATION
architecture consists of two subsystems: the smart garbage To provide a smooth and easy way for households to annotate
bin (SGB), embedded with distance and weight sensors to garbage content they dispose of daily. We further present

VOLUME 11, 2023 13003

 E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

FIGURE 1. Smart garbage bin system architecture design.

switch between the languages. Also, the interface consists of

house numbers as an identification for the experimental data
collection.
Fig. 3 demonstrates the garbage annotation application
interface whereby vertically depicts 8 garbage categories (i.e.,
Kitchen waste, Meal garbage, Paper/softbox, Fabric/textile,
Plastic, Dust, Plant, and All others) and horizontally depicts
25 garbage contents identities (i.e., Food garbage, Edible
food, Sink basin, Kitchen waste bag, Unclean cup, Unclean
container, Unclean packages, Waste wood, Tissues, Mixed
FIGURE 2. Smart garbage bin system overview. Papers, Milk/Juice box, Masks, Clothes, Shoe, bag, Rub-
ber products, Disposable diapers, Plastic product, Toys, CD,
Cigarette ashes/stick, Vacuum cleaner, Plant and Others)
a garbage annotation mobile application (GAA). The GAA belonging to each category. The garbage annotation appli-
designed and installed in a handy smartphone made a sig- cation provides a guide knowledge that allows individual
nificant value consideration to household users by allowing households to smoothly select the type of garbage content
annotation in a more efficient and tailored way through a each time they dispose of garbage in the SGB from a handy
smooth interface. The selection of the garbage categories in smartphone fixed outside on top of the SGB cover. Then, data
our proposed study is based on the rules for separating and about the garbage category and its specific identity content
disposing of burnable garbage as provided in four random are sent to the cloud data server using a Wi-Fi network.
selected municipal’s pamphlets in Japan that explain the
garbage disposal rules described in Section II-A, including IV. DEPLOYMENT AND DATA COLLECTION EXPERIMENT
the city of Kashihara [13], Ikoma [14], Nara [15], Kyoto [16]. Herein we present the experimental setup and data collection,
Additionally, we conducted a short survey with fifteen (15) including datasets, the data preprocessing steps undertaken to
students living in the city of Ikoma and Nara for one week. build the garbage contents estimation model, and the methods
The survey participants were asked to annotate their daily adopted to address the study aims. This study was approved
burnable garbage disposal on paper. The annotation included by the Ethical Review Committee for Research Involving
the name of the garbage contents and the frequency of dispos- Human Subjects at the Nara Institute of Science and Tech-
ing of such garbage. Thus, by analyzing the survey results nology (Approval No.: 2020-I-16).
and the rules for disposing of the garbage from municipal
pamphlets, we established important categories of burnable A. EXPERIMENT AND PARTICIPANT INFORMATION
garbage with specific content identities for the mobile annota- We conducted the evaluation experiment from June to August
tions application. The garbage annotations application inter- 2022 in five households of heterogeneous characteristics in
face comprises the garbage categories and a menu with two the city of Nara, Ikoma, and Kyoto in Japan for 3-5 weeks.
languages, English and Japanese, giving users flexibility to We considered family size, type of family, age group, number

13004 VOLUME 11, 2023

E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

TABLE 4. Garbage annotation frequency found in house 1 to 5.

We considered a time stamp of 10-minute intervals from

the disposal time recorded by the annotation application to
calculate features for the particular label. The features include
maximum, minimum, and rate of change of the garbage
filling level, weight, temperature, humidity, and air quality.
At the same time, the label consists of 8 garbage categories
and 25 garbage identities. Thus, we obtained the total original
FIGURE 3. Garbage annotation application interface.
datasets of each house for both garbage categories and content
identities. Below are the rules used to merge the collected
TABLE 3. Information of participants. data;
1) Every 10 minutes, if a new garbage label is input, and
then calculate new features for the label.
2) If at the same time or in less than 10 minutes, another
new label is input, then use the previously calculated
features for the new label (Overlap features).

C. CLASS IMBALANCE
A lower frequency of disposing of a particular type of
garbage content than the others experienced in all houses
of children, and city as the criteria for selecting participants leads to a minority of such garbage content. Therefore, the
for the experiment. Table 3 outlines the participant’s infor- minority class labels affect the model-building process, i.e.,
mation. All participants were well informed about the exper- a model that always chooses the majority class regardless
iment and provided their own consent to participate in the of the corresponding feature. To solve this, we utilize the
experiment. In addition, smart garbage bins were distributed resampling technique to enhance the classifier model’s size
and installed in each house. Fig. 2 shows the overview of the and quality and avoid biases class during training. There are
deployed SGBS. two main approaches for random resampling: Oversampling,
which duplicates the minority class, and Undersampling,
B. DATASETS which deletes the majority class. In our case, due to the low
The experiment resulted in five garbage logs data from number of annotations in garbage category 4 (Fabric/textile),
the five households. The garbage log consists of data from garbage category 5 (Plastic), garbage category 6 (Dust), and
the SGB (i.e., timestamp, filling level, weight, temperature, garbage category 7 (Plant) experience in all five houses (see
humidity, and air quality), collected every one-minute inter- Table 4), we applied the Oversampling technique to increase
val. Also, data from the GAA (i.e., timestamp, garbage cat- the minority class using the imbalanced-learn sci-kit-learn
egories, and content identities) collected only when a user library. Table 5 and Table 6 show the total number of datasets
disposes of and annotates the garbage in a smart garbage of garbage categories and content identities before and after
bin. The frequency of garbage disposal and annotation of resampling.
garbage contents differ in each household due to household
characteristics. Table 4 details the full annotations of garbage V. GARBAGE CONTENT ESTIMATION MODEL
contents found in houses 1 to 5 by the household users during This study aims to identify garbage contents disposed of
the experiment. Therefore, we define the following rules to and perform the garbage classification from garbage con-
merge the multiple sensor data from the smart garbage bin tents disposed of daily in the household by adopting IoT
(as features) and garbage content annotations by the house- and data-efficient machine learning algorithms. Therefore
holds (as labels) to create a single dataset of each house. we present a garbage content estimation model to classify

VOLUME 11, 2023 13005

 E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

TABLE 5. Re-sampling and cross-validation split for the 8 garbage first, we utilize repeated k-fold cross-validation to evaluate
categories.
the machine learning models in steps 1 and step 2 (see Fig. 4).
Then, we averaged the results with 4-fold cross-validations to
compute the final validation score for each investigated model
configuration. Therefore, the model created in step 1 used
the original (unbalanced) datasets, i.e., before resampling
(see Table 5). While the model developed in step 2 used the
balanced class dataset, i.e., after resampling (see Table 6),
as discussed in Section IV-C. Thus, for performance compari-
son of balanced and unbalanced datasets, our model-building
process output two models, an unbalanced model and a bal-
TABLE 6. Re-sampling and cross-validation split for the 25 garbage
content identities. anced model (see Fig. 4).
Afterwards, for better comparison reasons of the
cross-validation methods applied to the classifiers, and,
in order to increase the training set, in step 3 (see Fig. 4),
we changed the cross-validation method to leave one house
out cross-validation method where we repeatedly trained our
models with total balanced datasets from the four houses
and testing the model with the remaining one house. Thus,
we obtained the Leave one house out model.
Furthermore, we built the overall result models in step 4
(see Fig. 4) of the classification tasks for both class garbage
8 categories of garbage and a total of 25 garbage contents categories and content identities for each house to investigate
identities relating to a particular category, as demonstrated in the overall performance of the classifiers. We first made the
Fig. 3 of the garbage annotation application. The subsequent overall result model on all 8 garbage categories, i.e. Kitchen
section details the process of building classification models. waste, Meal garbage, Paper/softbox, Fabric/textile, Plastic,
Dust, Plant, and All others found in House 1, House 2,
A. MODEL BUILDING House 3, House 4 and House 5. Nonetheless, because each
Fig. 4 demonstrates model building steps and order of oper- garbage category comprises 5 to 2 specific garbage content
ations.Below we give a details explanation of the importance identities (see Fig. 3), in total, there are 25 different garbage
of each model-building step. We performed the classification content identities belonging to the eight categories expected
tasks from daily collected fuse sensor readings combined to be annotated by the users daily using the garbage annota-
with detailed household garbage contents annotations intend- tions application. Therefore because of the majority number
ing to find the class (i.e., 8 garbage categories: Kitchen waste, of garbage content identities and differences in frequency
Meal garbage, Paper/softbox, Fabric/textile, Plastic, Dust, behaviour of garbage disposal and annotation exhibited from
Plant, and All others) and (i.e., 25 garbage content identi- each house (see Table 4). In this study, we first selected
ties: Food garbage, Edible food, Sink basin, Kitchen waste the five garbage content identities from the Kitchen waste
bag, Unclean cup, Unclean container, Unclean packages, (category 1) as it has had a higher frequency of annotation in
Waste wood, Tissues, Mixed Papers, Milk/Juice box, Masks, house 3, house 4 and house 5. Also, we chose the five garbage
Clothes, Shoe, bag, Rubber products, Disposable diapers, content identities from the paper/softbox (category 3) as it has
Plastic product, Toys, CD, Cigarette ashes/stick, Vacuum had a higher frequency of annotation in house 1 and house 2 to
cleaner, Plant and Others) to which a new unseen observation learn the performance of the classifiers on garbage content
belongs. During the model-building steps in Fig. 4, we only identities. Therefore, to this point of the study, we created
consider utilizing data-efficient methods, namely: Random three overall result models for garbage content estimation,
forest, Naive Bayes, Extreme Gradient Boosting (Xgboost), namely;
and Decision tree algorithms to build the garbage content esti- 1) Overall result model for general garbage categories
mation model, for the reasons such as the comparison of the 2) Overall result model for kitchen waste contents identi-
machine learning classifiers, the small number of available ties
datasets, the popularity of the classifier and data preprocess- 3) Overall result model for paper, softbox contents identi-
ing to avoid minority class labels. We eventually defined the ties
order of operations applied to the selected classifiers during
the model-building steps.
More precisely, we train and test by spliting the dataset of B. PERFORMANCE EVALUATION
each house into four (4) chucks of 25% equal size dataset as Our model evaluation performance is based on accuracy,
shown in the Table 5 and Table 6 for garbage categories and which is the percentage of correct comparison classifica-
content identities. To avoid overfitting as much as possible, tions. Moreover, we evaluate the performance of our models

13006 VOLUME 11, 2023

E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

FIGURE 4. Model building steps and order of operations.

TABLE 7. 4-fold cross-validation performance accuracy for the 8 garbage categories.

TABLE 8. Leave one house cross-validation performance accuracy for the C. RESULTS
8 garbage categories.
Throughout this subsection, we describe results obtained
from the classification tasks as detailed in Section V-B.
Specifically, we look into and compare the performance accu-
racy from the unbalanced, balanced, leave one house out,
and overall result models using the four machine learning
classifiers.

1) UNBALANCED MODEL
We see from the results of the unbalanced model (see Table 7)
and (see Table 9) using the 4-fold cross-validations that
using other metrics, such as Confusion matrices, Precision, Random forest performs slightly better than other classifiers
Recall and F1-score. We will especially give the most infor- (Naive Bayes, Xgboost, and Decision tree), for classification
mative metrics for the overall result models because they tasks of both garbage categories and garbage content identi-
aggregated the garbage class label results from all houses ties. For garbage categories, the highest accuracy was 90%
belonging to the same classification and averaged the result obtained in house 1, and the 67% lowest accuracy resulted
into a single metric measurement. Furthermore, the model from the Decision tree in the same house. Also, 93% for
parameters tuning was applied on all classifiers, Random garbage content identities was the highest accuracy found in
forest, Naive Bayes, Extreme Gradient Boosting (Xgboost), house 1 by Random forest, and the lowest accuracy was 80%
and Decision tree. As a result, the accuracy slightly increased by the Decision tree found in house 4.
by increasing the number of parameters such as estimators,
criterion, and random state for each model separately. There- 2) BALANCED MODEL
fore, we independently investigated the model performance Afterwards, we compared the four classifiers with the same
on all experimental datasets found in House 1, House 2, 4-fold cross-validations method in all five houses on a bal-
House 3, House 4, and House 5 on garbage categories and anced dataset with the approaches discussed in Section IV-C
garbage content identities classification tasks. The percentage to deal with the unequal class balance. The results can be seen
performance accuracy results using 4-fold cross-validation in Table 7 and Table 9. We observed that the performance
and leave-one-house-out cross-validation as applied to the accuracy slightly decreased compared with the unbalanced
four machine learning classifiers for the 8 garbage categories model performance. Yet, Random forest manifested the high-
and 25 garbage identities are summarized in Table 7, Table 8, est accuracy and thus outperformed the rest of the classifiers.
Table 9, and Table 10. For the garbage categories, the Random forest exhibited 86%

VOLUME 11, 2023 13007

 E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

TABLE 9. 4-fold cross-validation performance accuracy for the 25 garbage content identities.

TABLE 10. Leave one house cross-validation performance accuracy for of garbage categories, (2) Overall result model of kitchen
the 25 garbage content identities.
waste contents identities and (3) Overall result model of
Paper/softbox contents identities. The performance accuracy
results for the three models are shown in Table 11. Moreover,
we compared the Recall, Precision, and F1-score for the
overall result models as they can better judge the performance
by showing the metric measurements of each class label.
For the garbage categories overall result model (see
Table 11), Random forest achieved the highest accuracy of
85%, followed by Naive Bayes at 82% and Xgboost at 80%,
in house 3, and 63% by the Decision tree in house 2 was while the decision tree lags with the least accuracy of 64%.
the lowest accuracy. While for garbage content identities, Table 12 summarises the metric accuracies of the 8 garbage
the accuracy was 88% by Random forest from house 1 and categories overall result model with Recall, Precision, and
house 2, and the most insufficient accuracy was 62% by a F1-score using the Random forest classifier.
decision tree in house 5. Further, for the overall result model of kitchen waste con-
tents identities (see Table 11) (i.e., food garbage, edible food,
3) LEAVE ONE HOUSE OUT MODEL sink basin, kitchen waste bag, and others). The Random
In the next step, we compare the results of the repeated 4-fold forest has steadily revealed the best classification accuracy of
cross-validation in step 2 to the Leave one house out (LoH) 91%, while the accuracies of the rest of the models are; 88%
cross-validation approaches in step 3 (see Fig. 4). In order to Naive Bayes, 84% Xgboost and 76% Decision tree. Likewise,
investigate the classification performance in all five houses. the overall result model of the paper/softbox contents identi-
Therefore, we applied the LoH on the balanced class datasets ties (see Table 11) (i.e., tissues, mixed papers, milk/juice box,
using the four classifiers in step 3. However, we maintained masks, and others) are 85% Naive Bayes, 83% Xgboost and
the same order of operation as in step 2. With this approach, 71% Decision tree were outperformed by the Random forest
the sum of four houses increases the size of the training at 89%. The summary of the Recall, Precision, and F1-score
set during repeated testing with only one house dataset. The for the overall result models of the 5 kitchen waste and the
results for Random forest, Naive Bayes, XGBoost, and Deci- 5 paper/softbox content identities are shown in Table 13 and
sion tree in the case of the garbage categories and garbage Table 14, using the Random forest as it has been portrayed as
content identities for all four classifier sets are shown in the best classifier.
Table 8 and Table 10. We see an apparent accuracy increase The aggregated confusion matrix plots using the Random
in each house compared to the balanced model of 4-fold forest of each overall result model are shown in Fig. 6,
cross-validation in Table 7 and Table 9. For the garbage where the columns represent the actual values (Truth) of the
categories, the Random forest revealed the highest accuracy target class label. The rows represent the predicted values
of 88% in house 3, while the decision tree showed the lowest (Predicted) of the target variable class label. The number of
accuracy of 57% in house 1. In addition, garbage content validation samples that were correctly classified are demon-
identities in the leave one house out model achieved the strated in the diagonal cells, and that were incorrectly classi-
highest accuracy of 91% and 90% by Random forest in house fied are demonstrated in the off-diagonal cells.
1 and house 2, respectively. On the other hand, the Decision In addition, to investigate the impact of the collected
tree exhibited unsatisfactory performance, 65% in house 5. multiple sensor readings on the garbage content estima-
Moreover, Random forest again steadily outperformed the tion model, we applied the features importance method
rest of the classifiers. using the Random forest classifier as our chosen classi-
fier for the garbage content estimation model. The results
4) OVERALL RESULT MODEL in Fig. 5 show that air quality, humidity, temperature,
To realize the performance of the three overall result mod- and fill level values are more relevant features for iden-
els described in Section V-A above Overall result model tifying garbage content in the smart bin. Therefore, the

13008 VOLUME 11, 2023

E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

TABLE 11. Accuracy performance of the three overall result models.

TABLE 12. Summary of 8 garbage categories overall result model.

FIGURE 5. Features importance on multiple sensor readings.

TABLE 13. Summary of 5 Kitchen waste contents identities overall result the highest accuracy is between 85% and 91%, and the low-
model.
est is 64%, which is satisfactory for garbage content clas-
sification tasks. However, the lowest amount of annotation
on certain class (imbalance) labels makes the classification
task difficult. We start the detailed discussion by compar-
ing garbage annotations from each house and then classifi-
cation tasks by the machine learning algorithms, followed
by the usefulness of the garbage content estimation model.
Finally, we look at the comparison of our approach to the
literature.
TABLE 14. Summary of 5 Paper/softbox contents identities overall result
model. A. COMPARISON OF HOUSEHOLD GARBAGE DISPOSAL
ANNOTATION AND CLASSIFICATION
In general, we observed different behaviour of garbage dis-
posal in all five houses, which is due to the heterogeneity
behaviour in each family, such as living style, size of the fam-
ily, type of the family, number of children/infants, age group,
and city. In this case, the study observed differences in the
routine frequency of garbage disposal and the type of garbage
content disposed among the houses. Therefore, using the
smooth garbage annotation interface (see Fig. 3) that allowed
identified garbage content disposed of daily and annota- household users to annotate garbage contents during disposal,
tion procedures contributes to the garbage classification the study found that certain garbage contents were important
tasks. Furthermore, the cross-validation approaches provided in some houses, i.e., daily disposed and annotated, com-
satisfactory results, especially for the leave-one-house-out pared to others. Table 4 shows the annotation frequency of
cross-validation, which performed better than the 4-fold garbage category disposal among houses, as briefly detailed
cross-validation. below.
• House 1: as shown in Table 3, this house consists of
VI. DISCUSSION a married couple in Kyoto prefecture. In this house,
Throughout this section, we discuss our findings and possible garbage category 3 (Paper/softbox) was the most impor-
implications. Due to the sufficient classification outcomes, tant category compared to other categories annotated
we chose the Random forest algorithm as the best classifier. 374 times during the experiment (see Table 4). In com-
We also decided on the overall result models as the final parison, garbage category 5, which consisted of plastic
model for our garbage content estimation tasks. Generally, contents, appeared as the least important annotated only

VOLUME 11, 2023 13009

 E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

FIGURE 6. Confusion matrices of the three overall result models.

5 times. In addition, other categories had almost a similar annotated only once each. Moreover, category 7 (Dust)
frequency of annotation, such as Kitchen waste (78), was not annotated in this house.
Meal garbage (66), All others (74), and Dust (50). On the • House 5: This house comprises a young married couple
other hand, fabric/textile had 21 annotations, while the with an infant in Ikoma city (see Table 3). Contrary to
plant had 19 annotations. all other houses, the study observed a fewer annotation
• House 2: consists of a married couple with two chil- frequency of garbage category 3 (Paper/softbox), which
dren living in Nara city (see Table 3). Like in house 1 prevailed in houses 1, 2, 3, and 4 as the most important
(see Table 4), garbage category 3 (Paper/softbox) was garbage category (see Table 4). Instead, kitchen waste
the most important category in this house, annotated was the most important category in this house, with
200 during the experiment, and Category 5 (Plastic) was 152 annotations, followed by Meal garbage (135) and
the least annotated, only 4 times. Compared with other Fabric/textile (77) third in the ranks. The high anno-
categories, Kitchen waste had 37 annotations, Meal tation frequency of category 4 (Fabric/textile) was due
garbage 63, All others 24, Fabric/textile had 16, dust to the disposal frequency of disposable diapers (the
11, and Plant 9. House 2 had fewer annotations than fourth garbage content in the Fabric/textile category
house 1. 4 see Fig. 3) thus increasing the number of fabric/textile.
• House 3: as shown in Table 3, this house comprises On the other hand, Plant category 7 was annotated only
a young married couple in Ikoma city. Even though once and therefore appeared as a minor category, similar
garbage category 3 (Paper/softbox) is steady as the most to house 3. Plastic had 9 annotations, and dust had
important and Plastic as the minor category observed in 6 annotations.
houses 1 and house 2, in this house, the study observed Eventually, daily disposed garbage contents and detailed
a slight difference in annotation frequency exhibited garbage annotation frequency by households impacted the
among Kitchen waste, Meal garbage, and Paper/softbox classification tasks in each house. For instance, in Random
categories. The result in Table 4 shows that the annota- forests, the chosen classifier for this study (see Table 7)
tions frequency kept, such as Paper/softbox (183), was and (see Table 9), the accuracies for classification tasks of
the most important, followed by Meal garbage (125), both garbage category and content identities in house 1 were
and Kitchen waste (104) was the third in the garbage higher than in house 4, which had fewer annotations frequen-
category importance ranking. cies. Moreover, the study found that the Decision tree was
• House 4: While Houses 1, 2, 3, and 5 comprise married the insufficient classifier model compared to Random forest,
couples, house 4 consists of two singles living in a shared Naive Bayes, Xgboost applied on the datasets in all five
house in Ikoma city (see Table 3). The study observed houses. Over and above that, the leave-one-house-out cross-
less annotations frequency in this house than in other validation method showed better performance compared to
houses. However, similar to houses 1, 2, and 3, garbage the 4-fold cross-validation approach despite its computational
category 3 (Paper/softbox) had the highest annotation cost (see Table 8 and Table 10). Therefore, in the overall
frequency and ranked as the most important, while the result models, we aggregated the classification result of the
plastic was minor. Therefore, the annotation frequency same class label into one metric performance using the leave-
in Table 4 is as follows: Paper/softbox had 61 annota- one-house-out approach, which has manifested better perfor-
tions, followed by Kitchen waste (23) and Meal garbage mance than 4-fold cross-validation on the balanced model.
(11), which similarly ranks with house 3. In addition, The following section compares our approaches with the
not only Plastic was the minor but also dust which was literature.

13010 VOLUME 11, 2023

E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

B. COMPARISON WITH LITERATURE C. STUDY LIMITATIONS

As discussed in the Section II, similar approaches in other • Few numbers of annotation
domains/applications were investigated. As detailed below, Our study provided sufficient burnable garbage iden-
we compare our strategies and experimental setups with those tification to guide house users during garbage dis-
more similar to ours. posal through the mobile application interface. Yet,
few annotations were recorded on some garbage cat-
• Suitable practice for house garbage separation egories because of the difference in garbage disposal
Our study has considered the identification of daily behaviour exhibited in each house. For instance, the
disposed of garbage content and provided a satisfactory low number of plastic, dust, and plant categories in
garbage category suitable for burnable garbage sepa- houses 1, 2, 3, and 4 (see Table 4), therefore, were
ration practice for most families in Japan. However, removed during model building as they were affect-
Nnamoko et al. [5] and Mookkaiah et al. [22] inves- ing the performance accuracy. For this reason, more
tigated only two kinds of garbage, i.e., Organic and garbage annotation is required for additional training
recyclable, which is not enough for rational garbage data to ensure a robust garbage estimation in application
separation in houses. Likewise, apart from increasing the scenarios.
number of classes as demonstrated by Ziouzios et al. [6] • Learn correct annotation
and Sami et al. [24], to find respective garbage cate- Even though the study identified the frequency of
gories such as (kitchen waste, other waste, hazardous annotations for each category in every house, house-
waste, plastic, glass, paper or cardboard, metal, fabric, holds need to learn and remember to correctly annotate
and other recyclable waste). Yet these studies provided garbage in a category and contents, which can further
a small number and more generalizable garbage cate- improve the garbage classification tasks.
gories, which is not the best practice for proper house
garbage separation and can not fully solve the prob- VII. CONCLUSION
lem of profound implications for ecological balance and This study presented a new smart garbage bin system (SGBS)
threat to global sustainability, development, and human embedded with multiple sensors to identify the disposed
well-being. garbage content categories by households. First, we designed
• Use of daily garbage contents and experiment trans- and developed a smart garbage bin system (SGBS) architec-
parency ture comprised of the smart garbage bin (SGB) equipped with
Our study proposed to perform garbage content esti- temperature, humidity, gas, ToF, and load cell sensors.Then,
mations from the daily collected fuse sensor readings we developed the garbage annotation mobile application
and household annotations with transparency on exper- (GAA) consisting of a smooth interface of 8 garbage cate-
iments and thus can be reproducible in the field. On the gories and 25 content identities to allow users to annotate
contrary, the studies by [6], [22], and [24] used pub- garbage contents during garbage disposal. Finally, we intro-
licly available garbage image datasets to improve clas- duce a new garbage content estimation method by train-
sification tasks with less transparency information on ing a machine learning model using daily collected fuse
their experimental setup. However, the publicly avail- sensor readings combined with detailed household garbage
able image datasets are associated with problems such contents annotations to perform garbage classification tasks.
as resizing, resolutions, and inappropriate colour pre- We deployed the designed SGBS in five households over one
sentation, thus lowering the quality of the classification month and applied the leave-one-house-out cross-validation
task. to the model trained and tested with the collected data. As a
• Use of efficient data models result, our proposed method achieved an accuracy of 91% in
Our study applied more data-efficient methods, namely 5 kitchen waste contents, 89% in 5 paper/softbox contents,
Random forest, Naive Bayes, Xgboost, and Decision and 85% in 8 garbage categories for the classification tasks.
tree, for the classification tasks. On the contrary, most of Moreover, our results show that air quality, humidity, temper-
the previous works applied the existing standard mod- ature, and fill level values are more relevant features in the
els for the classification tasks, such as VGGNet [17], garbage content estimation model.
AlexNet [18], ResNet [19], and DenseNet [20]. A com- The proposed SGBS contributes to household garbage
mon issue associated with image classification using identification and classification to ensure that valuable mate-
the existing standard model is high computational cost rials are recycled and utilized.
which often results in high development time and pre- Our future work includes extending our design to
diction model size because they are often pre-trained an event-based detection system to understand household
for more than one purpose [5]. In addition, CNN-based garbage disposal behaviour. Also, expansion of the experi-
models are difficult to run on embedded systems suit- ment to more families and experiment with other types of
able for garbage bins, and their architecture requires garbage, such as non-burnable garbage (Metal, glass, ceram-
large amounts of data for training which is yet to be ics, pottery, etc.) and recyclable (Plastic bottles, container
available. jars, cans, newspapers).

VOLUME 11, 2023 13011

 E. Likotiko et al.: Garbage Content Estimation Using Internet of Things and Machine Learning

REFERENCES [22] S. S. Mookkaiah, G. Thangavelu, R. Hebbar, N. Haldar, and H. Singh,

[1] What a Waste 2.0. Accessed: Nov. 5, 2022. [Online]. Available: ‘‘Design and development of smart Internet of Things–based solid waste
https://datatopics.worldbank.org/what-a-waste/ management system using computer vision,’’ Environ. Sci. Pollut. Res.,
[2] Ellen MacArthur Foundation. The New Plastics Economy: Rethinking vol. 29, pp. 64871–64885, Apr. 2022.
the Future of Plastics and Catalysing Action. Accessed: Nov. 5, 2022. [23] Trashnet. Accessed: Nov. 5, 2022. [Online]. Available: https://github.com/
[Online]. Available: https://ellenmacarthurfoundation.org/the-new- garythung/trashnet
plastics-economy-rethinking-the-future-of-plastics-and-catalysing [24] K. N. Sami, Z. M. A. Amin, and R. Hassan, ‘‘Waste management using
[3] The Global E-Waste Monitor 2020: Quantities, Flows and the Circu- machine learning and deep learning algorithms,’’ Int. J. Perceptive Cognit.
lar Economy Potential. Accessed: Nov. 5, 2022. [Online]. Available: Comput., vol. 6, no. 2, pp. 97–106, Dec. 2020.
https://https://collections.unu.edu/view/UNU:7737
[4] B. Wang, M. Farooque, R. Y. Zhong, A. Zhang, and Y. Liu, ‘‘Internet
of Things (IoT)-enabled accountability in source separation of household
waste for a circular economy in China,’’ J. Cleaner Prod., vol. 300,
Jun. 2021, Art. no. 126773.
[5] N. Nnamoko, J. Barrowclough, and J. Procter, ‘‘Solid waste image classi-
EUNICE LIKOTIKO received the B.E. degree from
fication using deep convolutional neural network,’’ Infrastructures, vol. 7, Ardhi University, the United Republic of Tanzania,
no. 4, p. 47, Mar. 2022. in 2012, and the M.E. degree from The Nelson
[6] D. Ziouzios, D. Tsiktsiris, N. Baras, and M. Dasygenis, ‘‘A distributed Mandela African Institute of Science and Technol-
architecture for smart recycling using machine learning,’’ Future Internet, ogy (NM-AIST), the United Republic of Tanzania,
vol. 12, no. 9, p. 141, Aug. 2020. in 2017. She is currently pursuing the Ph.D. degree
[7] C. Wang, J. Qin, C. Qu, X. Ran, C. Liu, and B. Chen, ‘‘A smart municipal with the Graduate School of Science and Tech-
waste management system based on deep-learning and Internet of Things,’’ nology, Nara Institute of Science and Technol-
Waste Manage., vol. 135, pp. 20–29, Nov. 2021. ogy (NAIST), Nara, Japan. Her research interests
[8] M. Cubillos, ‘‘Multi-site household waste generation forecasting using a include ubiquitous computing, smart home, and
deep learning approach,’’ Waste Manage., vol. 115, pp. 8–14, Sep. 2020. sensing technology.
[9] E. Likotiko, S. Misaki, Y. Matsuda, and K. Yasumoto, ‘‘SGBS: A novel
smart garbage bin system for understanding household garbage disposal
behaviour,’’ in Proc. 13th Int. Conf. Mobile Comput. Ubiquitous Netw.
(ICMU), Nov. 2021, pp. 1–8.
[10] Managing Municipal Solid Waste 32 European Countries EEA Report YUKI MATSUDA (Member, IEEE) was born in
no 2/2013ISSN 1725-9177. Accessed: Nov. 5, 2022. [Online]. Available:
1993. He received the B.E. degree in advanced
https://www.eea.europa.eu/publications/managing-municipal-solid-waste
course of mechanical and electronic system engi-
[11] W. Guo, B. Xi, C. Huang, J. Li, Z. Tang, W. Li, C. Ma, and W. Wu, ‘‘Solid
waste management in China: Policy and driving factors in 2004–2019,’’ neering from the National Institute of Technol-
Resour., Conservation Recycling, vol. 173, Oct. 2021, Art. no. 105727. ogy, Akashi College, Japan, in 2015, and the M.E.
[12] Japanese Waste Management and Recycling Industry. Accessed: and Ph.D. degrees from the Graduate School of
Nov. 5, 2022. [Online]. Available: https://www.env.go.jp/en/index.html Information Science, Nara Institute of Science and
[13] Separation and Disposal of Garbage in Kashihara Separation and Dis- Technology, Japan, in 2016 and 2019, respectively.
posal of Garbage in Kashiha. Accessed: Nov. 5, 2022. [Online]. Available: Since 2019, he has been an Assistant Professor
https://www.city.kashihara.nara.jp/documents/5c34c0f2f1a7f00f31b18cc1 with the Ubiquitous Computing Systems Labora-
[14] Garbage Collection Schedule in Ikoma City. Accessed: Nov. 5, 2022. tory, Graduate School of Science and Technology, Nara Institute of Science
[Online]. Available: https://www.city.ikoma.lg.jp/cmsfiles/contents/ and Technology. Since 2020, he has been a Researcher at Japan Science and
0000005/5895/gomical.pdf Technology Agency, PRESTO. His current research interests include partic-
[15] How to Divide and Take Out Garbage (for Household Garbage). ipatory sensing, location-based information systems, wearable computing,
Accessed: Nov. 5, 2022. [Online]. Available: https://www.city.nara.lg.jp/ and affective computing. He is a member of IPSJ, IEICE, JSAI, and ACM.
uploaded/attachment/34246.pdf
[16] Combustible Garbage. Accessed: Nov. 5, 2022. [Online]. Available:
http://kyoto-kogomi.net/wp-content/uploads/2016/04/gg-eng.pdf
[17] K. Simonyan and A. Zisserman, ‘‘Very deep convolutional networks for
large-scale image recognition,’’ 2014, arXiv:1409.1556.
KEIICHI YASUMOTO (Member, IEEE) received
[18] A. Krizhevsky, I. Sutskever, and G. E. Hinton, ‘‘ImageNet classification
the B.E., M.E., and Ph.D. degrees in information
with deep convolutional neural networks,’’ Commun. ACM, vol. 60, no. 2,
pp. 84–90, Jun. 2012. and computer sciences from Osaka University,
[19] K. He, X. Zhang, S. Ren, and J. Sun, ‘‘Deep residual learning for image Osaka, Japan, in 1991, 1993, and 1996, respec-
recognition,’’ in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), tively. He is currently a Professor with the Grad-
vol. 7, Jun. 2015, pp. 770–778. uate School of Science and Technology, Nara
[20] G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, ‘‘Densely Institute of Science and Technology. His research
connected convolutional networks,’’ in Proc. IEEE Conf. Comput. Vis. interests include distributed systems, mobile com-
Pattern Recognit. (CVPR), Jul. 2017, pp. 4700–4708. puting, and ubiquitous computing. He is a member
[21] Waste Classification Data. Accessed: Nov. 5, 2022. [Online]. Available: of ACM, IPSJ, SICE, and IEICE.
https://www.kaggle.com/datasets/techsash/waste-classification-data

13012 VOLUME 11, 2023