DETECTION OF TOMATO LEAF DISEASE IN LEAVES WITH DEEP LEARNING MOBILENETV2 WITH GAUSSIAN AND GABOR PREPROCESSING

Abstract

Global agriculture experiences significant economic losses together with diminished food security because of plant diseases that create threats throughout farming operations. Effective crop sustainability depends on early detection of plant diseases together with correct identification. The research introduces a time-efficient disease classification framework that utilizes BPPF and ACCF feature extraction methods together with Naïve Bayes and Decision Stump and JRip classifiers running on WEKA 3.9.5 platform. The evaluation standard utilized the Plant Village dataset consisting of 54,303 labeled images that covered 38 different classes of healthy and diseased leaves. A standardization process was applied to the input data through image resizing and normalization with additional augmentation techniques. The classifiers received training through extracted features from BPPF and ACCF while their performance evaluation used Accuracy, Precision, Recall and ROC-AUC, PRC-AUC along with Execution Time metrics. BPPF combined with Naïve Bayes classifier established 97.28% accuracy while ACCF and Naïve Bayes achieved 97.02% accuracy during 0.02 seconds of execution time. JRip-based models achieved high precision and recall numbers though their computational expenses remained high whereas Decision Stump models operated fast yet gave inferior classification accuracy results. Naïve Bayes classifiers operated with BPPF and ACCF descriptors reveal themselves as highly effective tools for real-time plant disease diagnosis in agricultural settings through their fast execution and interpretable method.