Enhanced Leaf Disease Detection Using Multi-Stage Segmentation and GAN-Based Classification Techniques

Abstract

 The primary source of financial loss in the global agriculture and farming sector is plant diseases. It’s a significant factor as it lowers crop quality and yield potential. Therefore, it’s imperative that various plant diseases be categorized and identified, and it demands the utmost care. Fruits are the vital nutrients rich source for plants worldwide, however the quantity and quality of fruits are negatively impacted by number of diseases. Results, early detection and classification of leaf disease are made possible by the application of efficient machine vision technologies.We put out a novel generative adversarial networkbased disease detection model for grape leaf. The industry is expanding positively. With an emphasis on limited grape leaf training images of illnesses, this study presents a unique model called Leaf GAN, which is based on generative adversarial networks (GANs), to generate pictures of four distinct disease images of grape for training of identification models. The deep connectivity approach and instance normalisation are integrated into an efficient discriminator model to identify realistic and fake images by using their excellent feature extraction capability on grape leaf lesions. Initially, a generator model framework with a decreasing count of channels is developed to generate diseased grape leaf images. Utilizing their advanced capability to capture details from grape leaf lesions, the approach employs dense connectivity in conjunction with instance normalization.The consistency and durability of the suggested method are also tested.After rigorous simulation, the proposed or developed model achieves a classified accuracy of 87.26% percent.The accuracy of the suggested work is better comparative to conventional machine learning techniques.