Autoencoder and GAN-aided plant disease detection in rice and cotton via hybrid feature extraction and decision tree classification

Abstract

In agriculture, crop diseases caused by pathogens, including bacteria, viruses, and fungi, pose a significant threat to the effectiveness of agricultural productivity. Some major crops in India such as rice and cotton are adversely impacted, leading to economic loss and loss of production. Timely intervention and sustainable agriculture depend on proper and early identification of diseases. In this paper, we propose a novel plant disease detection framework that integrates generative adversarial network (GAN) based image denoising with feature extraction and decision tree (DT) classification. The GAN module effectively removes noise from agricultural images, enhancing quality and stability under challenging imaging conditions. Following denoising, a combination of color, texture, and gradient features is extracted to obtain rich and discriminative patterns, which are then used to train a DT classifier for disease identification. Experiments are conducted on benchmark datasets comprising rice and cotton leaf images. The proposed system achieves superior performance, with 98.70% accuracy, 98.20% precision, 97.22% recall, and 98.50% F1 score, outperforming existing methods. These results demonstrate that the GAN-based denoising approach, combined with traditional feature-based classification, offers a robust, efficient, and practical solution for modern agricultural disease monitoring systems.