Recognizing the Plant Leaf Diseases Using the Transfer Learning Models

Abstract

Early and accurate detection can alleviate the impact of plant diseases while preventing large-scale damages that may cause a significant loss in crop production and quality. A big threat to global agriculture and food security is plant diseases. Conventional visual inspections have common characteristics such as being tedious, time-consuming, and suffering from human errors. We propose a system implementing a transfer learning technique (AlexNet, VGG16, and ResNet50) that allows for the automatic identification/classification of plant leaf diseases. The proposed model is trained and evaluated on a Plant Village dataset through a transfer learning process. The experimental results showed an outstanding classification performance on all three versions of CNNs are used. Resnet50 achieved the best result with Accuracy as high as 99.93%, while that of vgg16 reached to 99.87%, and AlexNet got down to 99.55%. It shows that the improved performance of ResNet50 demonstrates better effectiveness of deep residual connections in detecting minuscule pathological patterning located on leaves’ surfaces. This technique aims at promoting the sustainability of agricultural farming procedures. It reduces the dependence of human observation while maximizing crops yield and developing an automation tool that can diagnose all kinds of plant illnesses rapidly and easily.