See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/325107675

Machine Learning in Soil Classification and Crop Detection

Article · January 2016

CITATIONS READS

21 8,339

3 authors, including:

Rafega Beham

16 PUBLICATIONS 45 CITATIONS

SEE PROFILE

All content following this page was uploaded by Rafega Beham on 12 May 2018.

The user has requested enhancement of the downloaded file.

 IJSRD - International Journal for Scientific Research & Development| Vol. 4, Issue 01, 2016 | ISSN (online): 2321-0613

Machine Learning in Soil Classification and Crop Detection

Ashwini Rao1 Janhavi U2 Abhishek Gowda N S3 Manjunatha4 Mrs.Rafega Beham A5
1,2,3,4,5
Department of Information Science and Engineering
1,2,3,4,5
New Horizon College of Engineering, Bangalore, India
Abstract— This paper describes the SVM based classification restricts the development of agricultural wireless sensor
and grading of soil samples using different scientific features. network. Take soil moisture sensors for example, the existing
Different algorithms and filters are developed to acquire and methods to measuring soil moisture contain direct methods
process the colored images of the soil samples. These and indirect methods. Direct methods refer to those methods
developed algorithms are used to extract different features like to measuring the soil moisture by drying soil sample in order
color, texture, etc. Different soil types like red, black, clay, to estimate the exact soil moisture, which are standard
alluvial, etc are considered. The classification makes use of methods with great effort. Indirect methods refers to those
Support Vector Machine, machine learning technique. SVM methods to using other physical quantities related to soil
seeks to fit an optimal hyper plane between the classes and moisture including the neuron probe, gamma ray, resistance,
uses only some of the training samples that lie at the edge of time-domain reflect meter(TDR), microwave remote sensing
the class distributions in feature space (support vectors). This and so forth so as to estimate the actual soil moisture. Among
should allow the definition of the most informative training them, the neuron probe and gamma ray have high cost and
samples prior to the analysis. The accuracy of a supervised danger. High cost and complexity of soil moisture sensor
classification is dependent to a large extent on the training data become the bottleneck of promotion of agricultural wireless
used. Till now classification of soil and classification of crop sensor network. In this way, a soil moisture classification
for the appropriate soil is done separately. This project aims at model based on Support Vector Machine (SVM) with low-
combining both the techniques, where classification of crop cost sensors as a way of substitution for high-cost soil
for appropriate soil is a part of classification of soil. moisture sensors is proposed by the author to meet the
Key words: Crop Detection, Soil Classification practical needs of agricultural wireless sensor network.
In [2] author describes the prediction of yield rate of
I. INTRODUCTION crop, weather prediction, soil classification, crop
In past years, automation and intelligent sensing technologies classification and crop disease detection for agriculture
have revolutionized each and every part of our world like planning using statistics methods or machine learning
agriculture, building, aeronautics etc. These initiatives have techniques. This research proposed a technique named Crop
been accredited to the rising concerns about product quality Selection Method (CSM) to solve crop selection problems
and safety. Also, rising labour costs, shortage of skilled and to maximize the yield rate of crop over season and to
workers and the need to improve production processes have achieve maximum economic growth of the country in
all put pressure on producers and processors. Automated agriculture sector. Crop selector could be applicable for
solutions are the answer for the problems that are being faced minimize losses and to maximize crop yield rate when
today by the agriculture world. Agriculture planning plays a favourable conditions apply. There are two factors that
significant role in economic growth and food security of agro- influence the crop yield rate: first is seed quality which can
based country. Selection of crop(s) is an important issue for be improved using hybridization technology, second is crop
agriculture planning. Crop selector could be applicable for selection method based on favourable conditions. Crop
minimize losses when unfavourable conditions may occur production rate depends on geography of a region, weather
and this selector could be used to maximize crop yield rate conditions, type of soil, soil composition and harvesting
when potential exists for favourable growing conditions. methods. Based on different parameters different prediction
Machine Vision Systems (MVS) provide an alternative to models were developed. These prediction models are
manual inspection of soil samples for their characteristic classified into two different types: first is a traditional
properties and the amount of nutrient material. Computers statistics model where it generates a global model over entire
have been successfully employed for classification of plants, sample space. Second is machine learning technique which is
diagnosis of plant diseases, recognition of leaves and soil, etc. emerging technology for knowledge mining that relates input
Till now classification of soil and crops were done separately. and output variables which is hard to obtain statistically. IN
But in this methodology two techniques are implemented traditional statistics methods, structures of data model needs
together for providing a great usage for farmers and to be assumed priory, where as machine learning techniques
agriculture industry. need not assume this structure. This is useful characteristics
for machine learning techniques to model complex non-linear
behaviour in crop yield prediction. Machine learning methods
II. RELATED WORK
which are widely used in prediction technique are regression
In [1] author describes the measurement of soil moisture is an tree, support vector machine, k-nearest neighbours and
important issue in the research and application of agricultural artificial neural network. In this research, some machine
wireless sensor networks. The research on agricultural learning techniques are studied and comparative analysis is
wireless sensor network has practical and profound presented.
significance on the popularization of agricultural Internet of In [3] author describes an approach for automating
things and the realization of modern agriculture. However, classification procedure. Firstly, Segmentation algorithm is
the existing agricultural wireless sensor network still has the applied to segment the measured signals. Secondly, the
problem of high cost on agricultural sensors, which seriously salient features of these segments are extracted using

All rights reserved by www.ijsrd.com 792

 Machine Learning in Soil Classification and Crop Detection
(IJSRD/Vol. 4/Issue 01/2016/217)

boundary energy method. Based on the measured data and For more accuracy and clarity in the image its edges and
extracted features classifiers to assign classes with the contrast needs to be enhanced. Colour map of the image are
segments are built. This paper makes use of Decision trees, used for enhancing edges and contrast.
ANN and SVM. Cone Penetration Testing (CPT)
C. Segmentation
methodology is used for classifying sub-surface soil. Feature
extraction is performed using Boundary Energy Data by Once the enhancement of the image is completed in the
parameterising experts’ perception of the shape of the series previous stage using image pre-processing techniques
data. The overall classification scheme is based on CPT data. segmentation of the image is performed. Popular known
Soil Classification is done based on different algorithms for algorithm K-means Clustering algorithm is employed for
different soil types. In this paper a method to classify series segmentation of the image. It is used as a partition clustering
data where the constraint of contiguity has to be maintained which aims at partitioning a given data set into disjoint
is presented. subsets so that specific clustering criteria are optimized. The
most widely used criterion is the clustering error criterion
III. PROPOSED SYSTEM which for every point computes its squared distance from the
corresponding cluster centre and then takes the sum of these
A. Image Acquisition distances for all points in the data set. The centre of the cluster
Different images of soil samples which are to be classified are is picked and each pixel of the image is assigned to the
captured using color camera and are provided as an input to cluster. Re-computation of cluster centre is done by using the
the system. The features of each type of soil are collected and average of all the pixels. This iteration continues and the next
are stored in a separate database. This database is later used step is to take each point belonging to a given data set and
in the final stage for soil and crop detection. associate it to the nearest centroid.
K-means Algorithm is as follows-
1) Input: K, set of points x1........xn.
2) Place the centroids C1.........Ck at random locations.
3) Repeat until convergence:
For each point xi:
Fig. 1: Images of Soil Samples Find nearest centroid Cj.arg min D(xi,Cj)
Assign the point xi to cluster j
For each cluster j=1.....K:
1
Cj(a) = xi−> Cj ∑ xi(a) for each a = 1 … … d
nj
For each cluster j=1......K:
New centroid Cj= mean of all points xi assign to
cluster j in previous step.
4) Stop when none of the cluster assignments change.
D. Feature Extraction
After the segmentation of the image is performed by k-means
in the above phase our next step is the feature extraction stage.
Fig. 2: Flowchart of the proposed system
This is the foremost step in this methodology. All the features
B. Image Pre-processing that are required for us to classify the soil type and crop
The image acquired from our previous stage is not error free. detection are done in this phase. A number of features like the
The quality of the image is decisive for results of analysis as texture, colour, intensity, saturation, hue, etc are extracted for
it affects both the ability to detect features under analysis and detection of soil type. a filter known as Gabor Filter is
precision of subsequent measurement. In order to get an error implemented for feature extraction. Gabor Filter is a linear
free image pre-processing techniques are applied. This phase filter used for edge detection. Frequency and orientation
is also known as the enhancement of the image since the representations of Gabor filter are similar to those of human
image is enhanced by improving its contrast and removing visual system and they have been found to be particularly
errors to obtain a better quality image for our future appropriate for texture representation and discrimination. A
processes. The image contains errors like noise or artifacts set of frequencies and orientation representations may be
like scratches, lapping tracks, comet tails, etc which needs to helpful for extracting useful features from the image. Also
be eliminated before the further processes. Hence a filter other features like entropy, standard deviance, mean, etc can
called the Smoothing filter is made use to remove the noise be extracted using Gabor filter. The main and important
and artifacts from the image. There are two types of filters: feature of soil that is colour is needed to be extracted. Hence
low pass filter and high pass filters. Smoothing filter is a low a measure called colour moments are used to differentiate
pass filter. It is used to remove high spatial frequency noise images based on their features of colour. These provide a
from a digital image. Smoothing filter employs a moving colour similarity between images which can be compared to
window operator which affects one pixel of the image at a the values of images indexed in the data base for tasks like
time, changing its value by some function of a local region of image retrieval.
pixels. The operator moves over the image to affect all the E. SVM Classification
pixels in the image. Thus with many iterations the smoothing
Support Vector Machine (SVM) algorithm is used for soil
filter gradually enhances the image by removing the errors.
classification. It has successful applications in many fields

All rights reserved by www.ijsrd.com 793

 Machine Learning in Soil Classification and Crop Detection
(IJSRD/Vol. 4/Issue 01/2016/217)

like bioinformatics, text, image recognition, etc. SVM is a Used in Agricultural WSN”, Conf: on Information
universally accepted algorithm due to its simple nature. It is Technology and artificial Intelligence, Shanghai, China,
considered as an alternative to neural networks algorithm. pp.432-436.
The working nature is explained as follows: Given a set of [2] Kumar. R, Singh.M.P, Kumar.P, Singh.J.P, “Crop
training examples, each marked for belonging to one of two Selection Method to Maximize Crop Yield Rate using
categories, an SVM training algorithm builds a model that Machine Learning Technique”, (2015), Conf: on
assigns new examples into one category or the other, making ICSTM, pp.138-145.
it a non-probabilistic binary linear classifier. An SVM model [3] B. Bhattacharya, and D.P. Solomatine, "An algorithm for
is a representation of the examples as points in space, mapped clustering and classification of series data with constraint
so that the examples of the separate categories are divided by of contiguity", Proc. 3T"d nt. Conf: on Hybrid and
a clear gap that is as wide as possible. New examples are then Intelligent Systems, Melboume, Australia, 2003, pp.
mapped into that same space and predicted to belong to a 489-498.
category based on which side of the gap they fall on. SVM is [4] Soumya sree, Asst. Prof. A Rafega Beham, (2015),
effective to analyzing the separating planes and to identify the “Automated Diagnosis of Retina Images for Diabetic
largest margin so that the support to the data points will be Patients based on BP and SVM”, proceedings of the
identified. So we want to learn the mapping: X->Y, where ‘x’ International Journal of Computer Science and Mobile
belongs to X is some object and ‘y’ belongs to Y is a class Computing, NHCE, Bangalore, pp.299-306.
label.

Fig. 3: classification by SVM

This algorithmic approach is based on distinctive
characteristics analysis by analyzing the expected error
minimization. This approach considered the empirical risk to
improve the training procedure. The risk estimation is here
based on the structural analysis so that the generalization
error will be reduced. The error margin is analyzed under
class deviation and based on it nearest training patterns are
obtained. . This model is also based on the polynomial kernel
representation so that the effective learning to the elements
will be done and more accuracy will be obtained.

IV. CONCLUSION
The results from this study can be used for rapid identification
of soil types when they arrive in railcars at the terminal soil
elevators. Since this classification technique does not require
time consuming image processing routines such as Fourier
descriptors, it can readily be implemented using commercial
imaging libraries with Digital Signal Processing (DSP)
boards for real time operations. The work carried out has
relevance to real world classification of soils and it involves
both image processing and pattern recognition techniques.
SVM involves the sound theory first and then implementation
and experiments. A significant advantage of using SVM that
the solution to an SVM is global and unique, they have simple
geometric interpretations and gives sparse solution. The
computational complexity of SVMs does not depend on the
dimensionality of the input space and they are less prone to
overfitting.

REFERENCES
[1] Xiang Gao, Tancheng Lu, Peng Liu, Qiyong Lu,(2014),
“A Soil Moisture Classification Model Based on SVM

All rights reserved by www.ijsrd.com 794

View publication stats