Transfer Learning approach for smart agriculture: Application to Blight Bacteria

The presence of plant diseases causes considerable losses in global agricultural productivity. Most of these crop losses are due to diseases and weeds account for over 30% of global productivity. To avoid losses due to this problem, farmers around the world hire professionals to diagnose their crops. The traditional method of identifying these diseases by visually inspecting certain characteristics such as leaf color, shape, and texture is not always effective. However, this method is time-consuming and expensive. In this paper, we present an approach based on a transfer learning model to predict a plant’s disease by the symptoms of late blight, which include sudden and severe yellowing, browning, spotting, wilting, or death of leaves, flowers, fruits, stems or the whole plant. The plants used in this approach are potato, tomato, and rice. The models used in this study are ResNet50, VGG16, Inception V3, and MobileNet V2, of which MobileNet V2 has a better accuracy of 97.8%.