Proceedings of the Fifth International Conference on Communication and Electronics Systems (ICCES 2020)

IEEE Conference Record # 48766; IEEE Xplore ISBN: 978-1-7281-5371-1

An Analytical Approach for Soil and Land

Classification System using Image Processing

Prof. A. V. Deorankar. Ashwini A. Rohankar.

Department of Computer Science and Engineering, Department of Computer Science and Engineering,
Government College of Engineering, Government College of Engineering,
Amravati, India. Amravati, India.
[email protected] [email protected]

Abstract— In the last few decades researchers are interested model for classifying various kinds of soil series data along
in land mapping and its classification due to various reasons. The with a suitable suggestion for improving the fertility of the soil
reasons for an increase in the focus of the research community by detecting the health of the soil.
are, the increasing demand for agricultural land and soil health
analysis, as the health of the soil, is essential for the healthy A classification of the soil and identifying the quality level
production of crops. Image classification is one such approach for to which a soil belongs and what contents of the soil need to
soil and land health analysis. It is a complex process having the be improved can define the type of the soil. Knowing such a
effects of various factors. This paper has proposed the study of class or type of soil can be very useful for cultivation. For
current researches, the problems it addressed, and its prospects. analyzing the type of soil in a specific geographical area can
The emphasis is focused on the analytical study of vari ous be done by collecting soil samples of that area and using
advanced and efficient classification mechanisms and techniques. different machine learning algorithms classifying them into
Here, it has been attempted to study the factors these approaches various classes. With the emergence of machine learning and
have addressed to improve the accuracy of the classification. its implementation in image processing, the soil sample can be
Proper utilization of the number of features of remotely sensed classified efficiently into class to which it belongs. This paper
data and selecting the best suitable classifier are most important describes various studies and researches which have been used
for improving the accuracy of the classification. The knowledge - in the current domain.
based classification or Non-parametric classifiers like decision
tree classifier or neural network have gained more popularity for
multisource data classification in recent times. However, there is II. LIT ERAT URE SURVEY
still the scope of further research, to reduce uncertainties in the As discussed above, the number of researchers are working in
improvement of accuracy of the Image classification mechanisms.
this area. In paper [2], the authors observed that the number of
Keywords— Classification, soil, agriculture land, image processing, classification algorithms is available in remote sensing
machine learning methods like minimum distance, maximum likelihood, support
vector machine, K-NN, and multilevel classification. All the
classifiers work for the same – classification and accuracy.
I. INT RODUCT ION Edaphic factors are the external factors include the soil
moisture, soil air, soil mineral, soil temperature, soil organic
Soil possess different senses for different people like it is matter, soil organism, and soil reactions [3]. The growth of the
the products of past surface processes for a geologist. plant is completely depending on these factors. In our focus,
Similarly, for a penologist, it is a chemical and physical
organic matters are important because it provides the entire
process currently occurring. In India, Soil is particularly the
major, minor, and micronutrient to the plant. The role of the
basic and most essential entity for the agricultural Domain.
The product quantity losses can be reduced and the quality of organic component is improving the texture of the soil,
the crop can be improved if the characteristics of the soil can helping to increase
be recognized. It is very important for countries that have
several agricultural commodities to be export.
The water holding capacity of the soil. It is the food for most
Production of crops depends on four main factors like micro-organisms. It is observed that the growth of crops
climate, soil fertility, availability of water, and disease or depends on two factors as shown in figure- external factors
pests. And four biological factors as organic matter content, and internal factors.
Activation carbon content, Nitrogen content, and root health.
The health of soil can be tested in the rage of 1 to 100[1]. A
soil health test report provides an integrative assessment and
also identifies specific soil constraints. In this paper main
focus survey for soil health and different methodology used in
the classification process. This process will help in building a

978-1-7281-5371-1/20/$31.00 ©2020 IEEE 1416

Authorized licensed use limited to: Somaiya University. Downloaded on October 18,2022 at 14:22:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fifth International Conference on Communication and Electronics Systems (ICCES 2020)
IEEE Conference Record # 48766; IEEE Xplore ISBN: 978-1-7281-5371-1

that the classification is a part of the machine learning

mechanism.
Let's have a look at machine learning classification as shown
in figure 2. The machine learning approach is classified into
supervised and unsupervised learning method. The supervised
method is mostly used in the application. It has further
classified into classification and regression types.
Clustering is the type of Unsupervised Learning.

Figure 1 Types of Factors affecting to plant growth

Paper [4], works on the development of the vision system for

the soil where technology was based on image processing with
feature extraction. In this approach, images are considered as
input for classification. Paper also pointed out that the level of
nutrients will be helpful for farmers to come up with fertilizer
recommendations [5].

Another algorithm in vision séance is developed in [6] by A

Iriars and others. This group of researchers developed an
algorithm for ‘weed detection in the crop by computational
vision’. The purpose of this design was the management of
weeds in crops. A binary classification method was used in
Figure 2: Machine Learning Classification
this approach. As like [4], researches of this method also
consider an image as an input and perform the feature
classification with a binary method. Remote sensing-based classification uses the grouping of
pixels from the image which represents the land features in
case of land classification using image processing. In such a
In paper [7], researchers tried to implement such a type of
model for classification for agriculture purposes. The neural method, the land can be classified only with its application
such as agriculture land, forest land, or urban land. In image
network model for classification is used in many fields from
processing, different algorithms and methods have been used
biomedical to aeronautical engineering. Its application can
also observe in agriculture [7]. for such type of land classification. Few of them are
describing bellow-
In the next section, the basic structure of a few existing
methodologies. Each method has its merits and demerits. The fuzzy-based approach for classification is a supervised
learning approach and is commonly used in classification. The
Although the use of methods is going on and trying to improve
following figure shows the fuzzy-based approach from paper
the basic approaches and its area of application. In the
proposed approach, it has been tried to contribute to [10]. It has been seent that fuzzier is fixed between the inputs
and output level. Fuzzy system work on rules. Hence while
agriculture for detecting the health of soil and improvement.
designing a fuzzy logic system developers must be careful
about the set of rules and their values. Block diagram of a
III. EXIST ING M ET HODOLOGY fuzzy control system is as shown in figure-3

In the literature survey, a few research work and their brief

views have been considered. Now in this section, the existing
methodologies are studied. This is a broadly thinking view
about any classification system is that it should be easy to
handle and can handle all types of databases of soil. It should
also have deep knowledge of classification technically and it is
also said that the proper classification of features should be
done. And finally, it should be able to give maximum
accuracy.

Classification is the process of differentiating the content or

input based on their property. These properties may be internal
characteristics or it may functional characteristics or external
and behavioral characteristics [8] [9]. Simply it can be said

978-1-7281-5371-1/20/$31.00 ©2020 IEEE 1417

Authorized licensed use limited to: Somaiya University. Downloaded on October 18,2022 at 14:22:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fifth International Conference on Communication and Electronics Systems (ICCES 2020)
IEEE Conference Record # 48766; IEEE Xplore ISBN: 978-1-7281-5371-1

Figure 5: Neural Network classifier

Figure 3: Fuzzy Controller [*4]
K-NN is another approach in image processing used for
classification .basic block diagram of K-NN classifier is
The binary classification method is a very easy approach. shown in figure 6
When there are only two categories are available for
classification then the problem domain is said to be statistical
binary classification.

Figure 4: General classification in the binary classification Figure 6: K-NN Classifier

method

The neural network is the approach where, input needs to be IV. DEMERIT S OF EXIST ING A PPROACHES
preprocessed as extraction classification, segmentation, and  K-NN method is noise sensitive and high space
then finally proceed for the classification. In neural network complexity as well as high time complexity.
trained data is used for classification and pattern or feature  The neural network model is difficult to interpret and as
matching. In this case, it consists of three layers as input layer- K-NN it has also high space complexity and time
Hidden layered and output layer as shown in figure 5. complexity. Another disadvantage is its over-fitting[12].
 Space and time complexity are minima in binary
classification. However, the binary classification
approach has only two classes , and hence for multiple
classes, it is not a suitable method.
 The fuzzy approach is more accurate but as like other is
time-consuming and more complex designing with rule
base structure [11].
 The remote sensing method in image processing can be
used to a broad classification of lands like forest land,
agricultural land, or urban land. It is not suitable for
detection of nutrient and fertility level of the soil.

978-1-7281-5371-1/20/$31.00 ©2020 IEEE 1418

Authorized licensed use limited to: Somaiya University. Downloaded on October 18,2022 at 14:22:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fifth International Conference on Communication and Electronics Systems (ICCES 2020)
IEEE Conference Record # 48766; IEEE Xplore ISBN: 978-1-7281-5371-1

 In the case of Likelihood classification rate of confusion This database is later used in the final stage for soil
level or confusion matrix increases with increasing the classification. For successful image classification, a sufficient
allocation of high values to the pixel for classification. no. of training samples is required. The training samp les are
There is also a problem of overlapping distribution collected from fieldwork. The conditions considered while
associated with low- spatial resolution satellite images. selecting training samples included spatial resolution of the
 The minimum distance base classifier is only useful for collected images, availability of ground reference data, and
the shortest distance. complexity of the data being studied.
 SVM (Support vector machine) classifier is also another
popular classifier in this area. However, SVM has the
drawback of large space requirements and maximum time
of computation. That is time and space complexity is
again associated with the SVM classifier.

V. SCOPE OF T HE RESEARCH:
The government of India started a scheme ‘Soil Health
Card’ promoted by the Department of Agriculture & Co -
operation under the Ministry of Agriculture. It will be
implemented through the Department of Agriculture of all the
State and Union Territory Governments. An SHC is meant to
give each farmer soil nutrient status of his holding and advice
him on the dosage of fertilizers and also the needed soil
amendments, that he should apply to maintain soil health in
the long run.
A huge group of scientists is associated with land
classification, crop classification, and soil classification. There
is a need for classification of soil for its nutrient and its health. CONCLUSION
Few works are available associated with it. The proposed
research work can be introduced for classifying soil for its This paper surveys the different algorithms and methodologies
health detection. associated with the land classification and in this paper, it has
been attempted to identify a method for detecting the nutrient
All the methods that are involving with the training and testing level in the soil. Organic matters play a vital role in soil
phase have a limitation of the size of the database. It can be health. Uses of organic matters are good in séance of
worked on this factor to reduce the space complexity as well increasing water-holding capability and to provide major,
as time complexity. minor, and micronutrient to the plant. The good classifier
should handle diversity in the land. It should be hierarchical
From the survey, it has been observed that there are two major for deep classification with maximum accuracy. The level of
problems as- nutrients will be helpful for farmers for the further
 Lack of research on approaches and methodology in the recommendation of fertilizers. Fuzzy Logic with a rule-based
agriculture area for the detection and maintenance of soil system is highly modified and can perform more accurate
health. results of classification. On the other hand, binary
 The maximum existing algorithm has space complexity classification is a basic and fast approach, however, its
and time complexity at a high level. accuracy is low as compared to the fuzzy logic system
The primary aim is to work for agricultural soil and its health.
To find the best feature set from soil testing. References
VII. PROPOSED M ODEL [1] E. Ben-Dor et al., “Using imaging spectroscopy to study soil properties,”
Remote Sens. Environ., vol. 113, pp. S38 –S55, 2009.
Here a system is proposed for soil classification, which is [2] K.Kulkarni et al., ‘ Classification Using Remotely Sensed Data’
later described in the block diagram. The initial s egment is to International Conference on Computing Methodologies and
gather distinctive sorts of soil test picture which is regarded as Communication (ICCMC) Proceedings of the IEEE 2017
the selection of suitable sensor data is the first important step in [3] Factor affecting crop production-climate- edaphic pbiotic,Physiogaphic
image processing based soil classification as it requires and socioeconomic ,eagri.org › eagri50 › AGRO101 › lec09
considering factors such as users need, scale and characteristics [4] J carlo et,al. ‘Vision System for Soil Nutrient detection Using Fuzzy
Logic’ Proceedings of T ENCON 2018 - 2018 IEEE Region 10
of soils under study, the availability of data of soil, cost and Conference (Jeju, Korea, 28-31 October 2018)
time constraints of the study. Different images of soil samples [5] P. Rajasekharan, K. M. Nair, G. Rajasree and M. C. N. Kutty,
which are to be classified are captured using color camera and "SoilFertility Assessment and Information Management for Enhancing
are provided as an input to the system. The features of each Crop Productivity in Kerala," Kerala State Planning, 2013.
type of soil are collected and are stored in a separate database.

978-1-7281-5371-1/20/$31.00 ©2020 IEEE 1419

Authorized licensed use limited to: Somaiya University. Downloaded on October 18,2022 at 14:22:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fifth International Conference on Communication and Electronics Systems (ICCES 2020)
IEEE Conference Record # 48766; IEEE Xplore ISBN: 978-1-7281-5371-1

[6] A.Irias and etal; ‘Algorithm of weed detection in crop by computational International Symposium on computer vision and the
Vision’ CONIELECOMP 2019, 29rd International Conference o n Internet(VisionNet’15).
Electronics, Communications and Computing ©2019 IEEE [10] B. Bhattacharya, and D.P. Solomatine "An algorithm for clustering and
[7] Halimatu Sadiyah Abdullahi, Ray E. Sheriff, Fatima Mahieddine classification of series data with constraint of contiguity", Proc. 3T"d nt.
‘Convolution Neural Network in Precision Agriculture for Plant Image Conf: on Hybrid and Intelligent Systems, Melboume, Australia, 2003,
Recognition and Classification’ The Seventh International Conference pp. 489-498.
on Innovative Computing T echnology (INT ECH 2017) [11] Gordon, A.D. "A survey of constrained classification", Computational
[8] V. Sudharsan and B. Yamuna “Support Vector Machine based Decoding Statistics & Data Analysis, vol. 21, pp. 17-29, 1996.
Algorithm for BCH Codes” Journal of T elecommunication and [12] S. Haykin, Neural Networks: A Comprehensive Foundation, New
Information T echnology 2016. Jersey: Prentice Hall, 1999.
[9] Unmesha Sreeveni.U .B, Shiju Sathyadevan “ADBF Integrat able
Machine Learning Algorithms –Map reduce Implementation” Second

978-1-7281-5371-1/20/$31.00 ©2020 IEEE 1420

Authorized licensed use limited to: Somaiya University. Downloaded on October 18,2022 at 14:22:23 UTC from IEEE Xplore. Restrictions apply.