MACHINE LEARNING APPROACHES INCORPORATING EPCA AND ESVM FOR AUTOMATIC CLASSIFICATION OF PLANT LEAF DISEASE

Abstract

Farming ensures that all people will have enough to feed however if the world's population suddenly expands. It's also advisable to anticipate vegetation infections in their preliminary phase in the sector of agriculture is crucial to accommodate the fruits and vegetables to the wider public. Though it is problematic to identify the infections at the preliminary phase of the plants. The purpose of this study is to educate farmers on modern methods that may be used to lessen the prevalence of plant-leaf diseases. In this research, an automated "Leaf Disease Detection (LDD)" framework is developed with novel approaches of "Image Processing (IP)" and "Machine Learning (ML)" for identifying the type of diseases in the leaf. The proposed LDD uses a direct image of a leaf as its input. After the source image has been preprocessed to eliminate unwanted noise, the denoised image of the leaf was transmitted for the segmentation process where the "Region of Interest (RoI)" is segmented, then the segmented leaf image is fed into the feature extraction module. The primary objective of this Feature Extraction is to extract leaf characteristics from image data to transform it into a format that enables similarities amongst leaf images. In this research, we propose an "Enhanced Principal Component Analysis (EPCA)" for extracting the features from the segmented leaf image, that uses respectively statistical and directional features to identify disease categories in a leaf image. After extracting the features it moves on into the feature selection module to select optimal features, then the selected feature subset will move into the classification module. For classification, we propose an "Enhanced Support Vector Machine (ESVM)" framework with a "Weighting Kernel" to classify the leaf images into the following categories "Alternaria Alternata", "Anthracnose", "Bacterial Blight", "Cercospora Leaf Spot", and "Healthy Leaves". The performance tests for both the proposed "ESVM" classifier as well as the current "Bacterial Foraging Optimization based Radial Basis Function Neural Network (BRBFNN)" and "Advanced K-Nearest Neighborhood (AKNN)" classifiers have been evaluated. The parameters include "Accuracy", "Recall (Sensitivity)", "Precision", and "F-measure". Based on the findings, the proposed classifier outperforms the existing classifiers.

Keywords: LDD, PCA, SVM, AKNN, BRBFNN