Deep Learning for Robust Plant Leaf Disease Classification with Convolutional Neural Network

Biotic and abiotic stresses lower crop production in agriculture by 22 percent. These include diseases and environmental factors. Early identification of these stresses is very necessary for management purposes. Recently, computer vision techniques have turned out to be very effective tools in recognizing the first appearance of plant diseases. Deep learning techniques coupled with computer vision have given rise to recognition, either single or multiple, of the biotic stresses on a plant's leaf. It offers accurate segmentation of objects using a PlantVillage dataset. The architecture of a Hybrid-CNN is optimized for the segmentation of diseased plant leaves, having been trained with a combination of meticulous annotation and augmented data. Evaluations relative to traditional architectures like U-Net and Seg-Net imply that Hybrid-CNN models do better, much more so when it comes to instance segmentation. Semantic segmentation data help to identify and classify single leaf and multiple-leaf diseases with the highest level of accuracy. The newly proposed Hybrid-CNN model gives an impressive accuracy of 94 percent on the validation set, which consists of 10,861 images, thereby proving superior performance to modified segmentation models. These results strongly try to underline the efficacy of deep learning approaches in agricultural disease management.