Leaf Disease Detection using Neural Network Hybrid Models

Abstract

Around the globe, food plays a major role in the Ecosystem. Diseases caused by pathogens amount to a loss of 16% [1] of the annual yield. This, in turn, causes damage to economy as well. Advancements in Machine learning algorithms and Image processing techniques have made the detection of leaf disease comparatively easier and efficient than ever before. AlexNet is a CNN that has been considered one of the best for Image Classification. Other State of the Art (SOTA) [2] CNNs have also been considered in this paper while determining the best model for the problem at hand. The paper gives comparative analysis to classify 12 crop species with 38 leaf diseases using the CNN models considered effective for leaf disease detection. A hybrid approach presented here AlexNet+SVM gives a validation accuracy of 99.9986% with less than 0.01 error percentage. This is presented here and is proven to have outperformed some of the State of the Art (SOTA) CNNs like ResNet, Inception V3, VGG16, and AlexNet.