DEEP LEARNING-BASED COTTON PEST CLASSIFICATION USING COMPUTER VISION

Abstract

Images are used in agricultural sciences to monitor plant health and detect pests. This study addresses the issues of identifying cotton pests, specifically Whitefly, Jassid, Thrips, and Spotted Bollworm, using a proposed vision-based methodology that combines handmade and deep learning techniques. The project aims to extract statistical texture parameters from cotton field photos using 2D Gabor filters and co-occurrence matrices, such as smoothness, kurtosis, entropy, contrast, mean, and homogeneity. These features were refined by experimental pruning and classified using a Support Vector Machine (SVM) using a Fine Gaussian kernel. Performance was evaluated using metrics such as precision, recall, F-measure, ROC-AUC, mean absolute error, and root mean square error, resulting in a pest classification accuracy rate of 94.3%. This research makes a significant contribution to the development of automated and precise pest detection technologies, boosting sustainable agricultural practices, particularly in locations like Southern Punjab, Pakistan.