• Nu S-Au Găsit Rezultate

View of A Hybrid Model for Pest identification in Groundnut Plants using BLR and SVM Techniques

N/A
N/A
Protected

Academic year: 2022

Share "View of A Hybrid Model for Pest identification in Groundnut Plants using BLR and SVM Techniques"

Copied!
9
0
0

Text complet

(1)

A Hybrid Model for Pest identification in Groundnut Plants using BLR and SVM Techniques

Mr. K. Gowrishankar, M.C.A., M.Phil

1*

, Dr.S.LakshmiPraba, Ph.D

2

,V.Divya Darshini

3.

1Research Scholar, Department of Computer Science, Periyar University, Salem, Tamilnadu

2Assistant Professor, Department of Computer Science, Govt. Arts College for Women, Salem- 8, Tamilnadu.

3PG Student, M.Sc. Software System, Coimbatore Institute of Technology,Coimbatore, Tamilnadu.

*[email protected]

ABSTRACT

Plants and trees are vital for life of person and animal. The plants have diseases that are also caused by fungi, insects, Pest and viruses. For identifying the types of pests and the percentage of disease affected, various strategies have been developed and applied. Techniques for image classification are commonly used in a variety of applications.This paper investigates the use of a hybrid model of Binomial Logistic Regression and Support Vector Machine (BLR-SVM) for pest detection in groundnut leaf photographs. Image classification plays a vital part of pest identification by image processing.For pests and diseases classification, a range of techniques and algorithms have been developed. The efficiency of a hybrid method of BLR-SVM classification algorithms for detecting pests in groundnut leaf images is examined in this article. The Accuracy, F1-Score, and Area Under the Curve (AUC) are used to demonstrate the efficiency of the hybrid image classification method in the analysis.

Key words: Image Classification, Binomial Logistic Regression, Support Vector Machine.

1.Introduction

India is a beautiful country with a diverse range of physiographic and climatic environments. While this climatic condition is ideal for crops such as vegetables, fruits, flowers, nuts, and plantation crops, the quantity of crops is ruined each year due to unfavourable climatic conditions or pathogen incursion. These circumstances aggravate the case and cannot be ignored or eliminated permanently. The key threatening criterion for crops is the identification of disease in a plant leaf, which is one of the conditions listed above.

Plants are highly vulnerable to diseases due to their environmental exposure. The most crucial step is to detect diseases in developing plants at the earliest possible time. Owing to the shortage of labour and cost of fertilisers, manufacturing cost is getting higher in the current situation, these are the most significant consequences for farmers. In order to intensify the quality & quantity of goods it is important to a look for an upgraded technology.

With significant access, image analysis is a big methodology for parameter assessment in determining plant health-related issues. Image analysis is used to evaluate field input, diagnose disease development in a plant at an early stage, and prevent pests from spreading to other areas of the plant.Groundnut is one of the plants that is most afflicted by a variety of pests, so it is taken into consideration.

In the last decade, progress on image recognition and pest detection in plants has increased dramatically.

Groundnut leaf image processing is notable because of the variety of techniques used for pest identification. The automated leaf identification system is used to identify pests in leaves. It invigilates the vast fields and mechanically detects the individual pests. This paper discusses an approach to automated image treatment.Images from the field are initially obtained using some of the computing devices. Any computing technique can be used to extract vital data from an image. Unwanted pixels are removed from the obtained data using a filter. Multi threshold base colour segmentation technique is used to segment the filtered image.

Picture segmentation can also be thought of as a method of dividing a single image into several segments. It will smooth out the picture and make it easier to calculate. When combined with SVM classifiers, feature extraction is one of the roles to find the simplest results [1]. In colour image analysis, the segmentation problem is one of the most critical. Color uniformity can be used to divide a picture into substantial regions as a fundamental principle behind colour images segmentation.. Just some aspects of the photo are of concern to people. The identification of related regions uses properties such as strength, texture, depth, gray-level and colour support;

these qualities are used in group areas for good meaning.Segmentation is an important tool in many areas including health care, industry, remote sensing, image analysis, content-based images, pattern identification, traffic image, video and computer vision. Edge, cluster, threshold, regionally dependent, and hybrid segmentation technologies are some of the commonest techniques..Supported threshold segmentation of images is one of the longest, most efficient approaches as the edge value divides the pixels in such a way that the value of the pixels is of the intensity not of the threshold, and the intensity pixels belongs to a particular class[2,3].

(2)

One of the most basic segmentation methods is thresholding. To transform a grey image into a binary picture, this method uses a clip-level (or a threshold value). Balanced histogram thresholding is also available. Picking the edge value is the key to this process (or values when multiple-levels are selected). In the industry, many common methods are used, including the utmost enlightenment.For segmentation, supervised methods involve expert human input. Various clustering and classification algorithms, such as Fuzzy C Means and K Mean, were used in pest identification methods [4-6,9].Region-based approaches divide a picture into distinct regions that are similar based on a collection of predetermined criteria. In general, it means that the training data are selected by human experts, and then used to segment the images. Semi-automatic or fully automated methods are unmonitored[12][13]. The user interface can be necessary at some stage[14], but the outcomes should be more or less human autonomous, to increase the performance of the processes..

2.Literature Review

Several approaches and ID systems are designed to accomplish this purpose and are able to respond to very complicated and diverse types of images[7]. For the analysis and classification of the framed image, the MATLAB tool is used[8]. The objective of segmentation is to simplify and transform image representation into something easier to analyse and more meaningful. The classification algorithm uses the training data to predict whether an image belongs to one of the predetermined categories.In the Machine Vision area, the information collected from the photo is used to identify other activities, e.g. an airport for remote sensing.

Table I.On the Application, a Few Related Works

In this literature review, several methods of segmentation and classification were proposed. The segmentation technique is selected by the specific type of image and features of the issue to be examined over the segmentation stage. The classification technique predict accurately whether the given class is belongs to normal classes or abnormal classes.

3.Methodology

The study developed focuses on identifying and grouping noodle disease using the technique of image processing. For pests and diseases classification, a range of techniques and algorithms have been developed.This technique collects and feeds input photographs of the noodle for further study. The preprocessed image is then segmented by colour segmentation depending on certain thresholds.The performance of hybrid approach of BLR-SVM classification algorithms for detecting the pests in groundnut leaf images. The experiment results of the hybrid image classification aproach performance with the Accuracy, F1-Score,and Area Under the Curve (AUC). The Figure1. represents the proposed methodology.

YEAR AUTHOR METHODOLOGY APPLICATION USED

2021 Gayatri Pattnaik et al SVM with HOG and

LBP Pests identification in Tomato leaves 2020 K.Gowrishankar et al ANN and GLCM Diagnosis of disease in plant leaves 2018 Vijayalakshmi et al K Nearest Neighbour Identification of infected leaves 2018 Kumari et al Fuzzy C Mean Intelligence gathering in plant leaves of the

impacted area

2017 Jayamoorthy et al K Fuzzy C Mean Disease recognition and recommend Leaf pesticides

2017 Devaraj et al SVM Algorithm Bean cultivation-Identification of diseases

(3)

Figure 1. Work flow diagram for proposed work

Performance of Median Filter In Noise Reduction

For the picture that has no unique noise, median filters provide complicated math analysis. The noise variance is roughly below standard distribution for picture with zero medium noise.

𝜎2𝑚𝑒𝑑 = 1

4𝑛𝑓2(𝑛 )≈ 𝜎𝑖2 𝑛 +𝜋

2− 1.𝜋

2 − −(1)

The medium filtering efficiency is higher for random noise reduction than the average filtering performance.

The preprocessed image is then fed for further process segmentation.

Color Image Segmentation thresholding technology

The most basic image segmentation tool is thresholding. Thresholding is usually used for creating binary representations from a grey pixel. Thresholding is used in the image processing to divide the image into smaller fragments or junks using at least one colour or grey to establish the border. The benefit of first getting a binary image is that the ambiguity of the information is reduced and the process of identification and classification is simplified [2].

A grey image is most often converted into a binary image by selecting a single threshold value (T)[2].

The input to a threshold is normally a grey or colour picture. The performance can be a binary image representing the segmentation in the simplest implementation. The white hue is the front and the Black shades are the rear (or vice versa). This segmentation approach applies a fixed criterion simultaneously to all pixels in the image. The segmented picture is a separated representation with continuous areas or pixel sets. The pixels are divided by their value of strength. Image segment in front and background.

G(x, y) = 1 when f(x,y) is the first pixels = 0 when f(x,y). In actual implementations, histograms with multiple peaks and unclear valleys are more complicated and the value of T cannot be always selected easily.

Input leaf image

Preprocessing (Median Filter)

MultiThreshold based Colour Segmentation

Classification

10 fold cross validation

BLR SVM

Prediction model

Evaluation Finding Accuracy, F1- Score

(4)

---(2) The following method can be defined as: T=T[x, y, p(x, y), f(x, y)

Where f(x, y) meted the local property to be the gry level and p (x, y).

Where F(x, y) is larger than T, the object point is otherwise named the point is not a point of the object since it relates to the object's background.

Global Thresholding:

Many of the global technology thresholds are appropriate because there is a relatively distinct distribution of artefacts and background pixels. One threshold value is included throughout the whole picture at the global threshold. In a couple of years, the global threshold was a common strategy. The backdrop of the image with the certain pixel values was divided by the image values in the foreground and a global threshold could be used.

Global thresholding involves the selection of the T threshold in the foreground. If, at any global threshold, g(x,y) may be a threshold variant of f(x, y),

---(3)

Test methods such as: Otsu, optimum threshold analysis, histogram analysis, iterative thresholding, maximum threshold similarity, clustering, multi-spectral and multi-thresholding.

BLR (Binomial Logical Regression)

The Binomial Logical Regression is an algorithm-controlled model of computer education, The BLR provides the connection between the expected variable and the real. The likelihood of good events for the BLR model is the dependent variables and the chance of failure is the independent variables.

Where the class is a response vector and T, H variables, i.e. temperature, humidity, are projected. Then we can write the BLR equation as

---(4) The logic function is named as odds by the above equation and p(Class)1-p(Class).

Here, 0 to 2 is the regression coefficients to calculate the probabilities of class is belongs to 0 or 1.

Where, the inverse of above equation is given

---(5)

Here, the above function is the sigmoidal function and it gives the Possibilities ranging from 0 to 1.

Take into account that p is greater than 0.5 and round it as 1, it should be more than 0.

Support Vector Machine (SVM)

The Support Vector Machine or SVM is among the most significant common monitoring learning algorithms used for classification and regression problems. It is mostly used in machine learning for classification problems. The SVM Algorithm aims to construct the best line or judgement limit, which can

(5)

separate n-dimensional space into groups so that in future the new data point can conveniently be placed in the right group. It's a hyperplane that is the best judgement limit. In order to produce the hyperplane, SVM selects the extreme points and vectors. These extreme cases are known as vectors of support and are also called Support Vector machine.

Mesurements of performance by Confusion Matrix.

Classification model is very useful only if it gives a accurate results in the given problem.The measurements from the model makes a essential task for prediction by the researchers

Confusion Matrix: Confusion matrix is a tool with the right and wrong estimation calculation. The parameters of the confusion matrix is

TP - (True Positive): The classifier predicted the number of sample as positive class and it also belongs with the the positive class

TN – (True Negative):Classifier predicted the test amount as negative class and it also belongs with the negative class.

FP – (False Positive): Classifier predicted the test amount as negative class and it also belongs with the positive class.

FN – (False Negative):Classifier predicted the test amount as positive class and it also belongs with the negative class.

4.Experimental Results

In this section , performance of the experiment on the image dataset. This prediction model having SVM, BLR and Hybrid BLR-SVM model, the comparison done by using performance metrics of Accuracy and FI-Score. The Preprocessed image is further fed into image segmentation using mutithreshold image segmentation. The figure 2 shows the segmented image and original images of the leaves with the caterpillar.

Figure 2. The original images and segmented images of leaves with caterpillar.

The performance of Binomial Logistic Regression shown as a confusion matrix in the Table II. It shows the table between the predicted classes and actual classes

Table II. Confusion Matrix for BLR Classification

Actual: YES Actual: NO Predicted:

YES 50 6

Predicted:

NO 12 55

(6)

From the above table of confusion matrix shows the performance of BLR Classifier. The measuring of the performance by calculating Accuracy FI-Score using TP,TN,FP andFN.

Where

TP=52; TN=51; FP=4; FN=16

Accuracy=(52+51)/(52+51+4+16)=0.8373=83.73%

AUC=1/2*((52/(52+16))+(51/(51+4)))=0.8459 PPV= 52/(52+4)=0.9285

TPR=52/(52+16)=0.7647

F1-Score=2*(0.7647*0.9285)/(0.7647+0.9285)

=0.8386

Table III. Confusion Matrix for SVM Classifier

From the Table III. of confusion matrix shows the performance of SVM Classifier. Here the True Positive value have a slight increased when compared to BLR classifier. The measuring of the performance by calculating Accuracy FI-Score using TP,TN,FP andFN.

Where

TP=50; TN=55; FP=6; FN=12

Accuracy=(50+55)/(50+55+6+12)=0.8536=85.36%

AUC=1/2*((50/(50+12))+(55/(55+6)))=0.8540 PPV= 50/(50+6)=0.8928

TPR=50/(50+12)=0.8064

FI-Score=2*(0.8928*0.8064)/( 0.8928+0.8064)

=0.8474

Table IV. Probabilities of Class Labels With the Parameters.

The Table IV shows the probabilities of class labels based on the temperature and humidity parameters. The comparison for the predicted class is shows in the table V. Here, if the positive value is more than the negative value then the class value is to be 1 and otherwise 0.

Table V.Probability Comparison for Predicted Class

Comparison Class

If P>N 1

If N>P 0

From the above modified data makes a minimized data sample with a class value with the parameters value of Temperature and Humidity shows in table V.

Actual: YES Actual: NO Predicted:

YES 52 4

Predicted:

NO 16 51

Parameters Probabilities of class labels

T H P N

23.2 91 0.997 0.002

24.1 88 0.001 0.998

24.3 82 0.981 0.018

25.1 73 0.001 0.998

(7)

Table V. Data Samples for Minimized Variable.

The obtained train set was fed into SVM classifier for developing a new classification model for caterpillar detection. Hence the hybrid model BLR and SVM classifier makes a new prediction model for identifying pests in plant leaf. The confusion matrix created for the hybrid classifier is shown in table.

Table V1. Confusion Matrix for BLR-SVM Classifier

Where

TP=61; TN=51; FP=3; FN=9

Accuracy=(61+51)/(61+51+3+9)=0.9032=90.32%

AUC=1/2*((61/(61+9))+(51/(51+3)))=0.9079 PPV= 61/(61+3)=0.9531

TPR=61/(61+9)=0.8714

FI-Score=2*(0.9531*0.8714)/( 0.9531+0.8714)

=0.9104

From the above confusion matrix for the hybrid classification model is shows that the number of true positive value is higher than the SVM and BLR models. The performance measures of Accuracy, AUC, and F1-score can be calculates by applying the values to the concern equations The accuracy increased up to 94.32%. F1- Score of the hybrid model is 0.9426. The AUC for the proposed hybrid model is 0.9414 respectively.

Table VII Performance Metrices of Classifier Models.

Classifier model Performance Metrices

Accuracy F1-Score AUC

BLR 83.73% 0.8386 0.8459

SVM 85.36% 0.8474 0.8540

BLR-SVM 90.32% 0.9104 0.9079

From the above table of performance metrices of different classifier models compared with the proposed model using a graph shown in the fig. 2.

Parameters Value Class

(Dependent Variable)

T H

23.2 91 1

24.1 88 0

24.3 82 1

25.1 73 0

Actual: YES Actual: NO Predicted:

YES 61 3

Predicted:

NO 9 51

(8)

Fig.2. Performance of BLR, SVM and BLR-SVM Classifier.

5. Conclusion

Plants develop a variety of flaws over the course of their lives. They are the big and mandatory component of our surroundings. They are often afflicted by pests brought on by environmental influences and other climatic conditions, all of which have an effect on plant development. Therefore a method for the identification of plant leaf pests is developed which is obtained by automatic detection and classification of pests. The decrypted work uses BLR-SVM for colour transformation, function extraction and classification to achieve high efficiency rates.As compared to similar methods, the evolved approach outperforms them in terms of computational efficiency and accuracy. Other classification methodologies will be able to discern the developed work in the future.

6. References

[1] Gayatri Pattnaik, K.Parvathi (2021). “Automatic Detection and Classification of Tomato Pests Using Support Vector Machine Based on HOG and LBP Feature Extraction Technique”, Progress in Advanced Computing and Intelligent Engineering, 1199, 49-55.

[2] K.Gowrishankar, S. Lakshmi Prabha.(May 2020). “An Integrated Image Processing Approach for Diagnosis of Groundnut Plant Leaf Disease using ANN and GLCM”, Journal of Scientific & Industrial Research, 79, 372-376.

[3] S.Vijayalakshmi, D.Murugan (2018) “Comparative Analysis on Segmentation Approaches for Plant Leaf Disease Detection”, International Journal of Computer Sciences and Engineering, 6(5), 412-418.

[4] Monalisa Mishra et. Al.(2019) “ A Robust Pest Identification System Using Morphological Analysis in Neural Networks”, Periodicals of Engineering and Natural Sciences, 7(1).483- 495.

[5] A.Kumari, S.Meenakshi, S. Abinaya.(2018).” Plant Leaf Disease Detection Using Fuzzy C-Means Clustering Algorithm”, International Journal of Engineering Development and Research, 6(3), 157-163.

[6] M.Thilagavathi, S. Abirami.(2018).” Application of Image Processing in Detection of Plant Diseases: A Review”, International Journal of Research and Analytical Reviews, 5(1),403-406.

[7] S.Poonkuntran, M. Kamatchidevi, L. S. Poornima, R. Shreeja.(2018).” Plant Disease Identification system”, International Research Journal of Engineering and Technology, 5(3), 2245-2250.

[8] Sandesh Raut 1, Amit Fulsunge.(2017). “Plant Disease Detection in Image Processing Using MATLAB”, International Journal of Innovative Research in Science, Engineering and Technology, 6(6), 10373-10381.

[9] S.Jayamoorthy, Dr. N.Palanivel,(2017).” Identification of Leaf Disease Using Fuzzy C-MEAN and Kernal Fuzzy C-MEAN and Suggesting the Pesticides”, International Journal of Advanced Research in Science, Engineering and Technology,4(5), 3852-3855.

[10] Devaraj P, Megha P Arakeri and B.P. Vijaya Kumar. (2017). ” Early detection of leaf diseases in Beans crop using Image Processing and Mobile Computing techniques”, Advances in Computational Sciences and Technology,10(10), 2927-2945.

[11] Megha .S, Niveditha C, SowmyaShree .N,Vidhya.(2017) .K,” Image Processing System for Plant Disease Identification by Using FCM-Clustering Technique”, International Journal of Advance Research, Ideas and Innovations in Technology,3(2), 445-449.

0.8000 0.8200 0.8400 0.8600 0.8800 0.9000 0.9200

BLR SVM BLR-SVM

Accuracy F1-Score AUC

Classifiers

performance

(9)

[12] Batoo, Khalid Mujasam, et al. "Behavior-based swarm model using fuzzy controller for route planning and E-waste collection." Environmental Science and Pollution Research(2021): 1-15.

[13] Chang, Jinping, Seifedine Nimer Kadry, and Sujatha Krishnamoorthy. "Review and synthesis of Big Data analytics and computing for smart sustainable cities." IET Intelligent Transport Systems (2020).

[14] Song, Hesheng, and Carlos Enrique Montenegro-Marin. "Secure prediction and assessment of sports injuries using deep learning based convolutional neural network." Journal of Ambient Intelligence and Humanized Computing 12.3 (2021): 3399-3410.

Referințe

DOCUMENTE SIMILARE

Keywords:Artificial neural networks, Data mining techniques, Meteorological data, Rainfall prediction, Support Vector

Toate acestea sunt doar o parte dintre avantajele in care cred partizanii clonarii. Pentru a si le sustine, ei recurg la o serie de argumente. Unul dintre ele are in atentie

of red.uced lteimite interpolation to fill up cerlain gaps in the finite element construction of recfangular elements' It is the topic of a forthco-.. min he

10 “It is obligatory to take into account the cultural and political path dependencies of the countries and the secularization effects”, Detlef Pollack, Olaf Müller,

Thus, if Don Quixote is the idealist, Casanova the adventurous seducer, Werther the suicidal hero, Wilhelm Meister the apprentice, Jesus Christ will be, in the audacious and

(2020) proposed a new hybrid approach using different machine learning techniques to predict the heart disease.. Classification algorithms like Logistic Regression,

Here, a novel method is known as the Hybrid Linear stacking model for feature selection and Xgboost algorithm for heart disease classification (HLS-Xgboost)1. This model

had suggested machine learning based automatic segmentation and hybrid feature analysis for Diabetic Retinopathy classification using fundus Image (2020).. This study used