Deep Learning Architectures Extended from Transfer Learning for Classification of Rice Leaf Diseases

Rice is one of the world's five main food crops. The problem helps farmers identify diseases on rice leaves early and develop a plan to prevent diseases in time; at the same time, helping them reduce damage and increase crop yields is of great interest to the agricultural sector. However, with the cultivation on a large scale, the detection of rice diseases by experience or manual form is still limited. In recent years, the application of Deep Learning techniques to detect disease identification in rice through images has yielded many superior results compared to traditional methods. This study has leveraged and extended transfer learning convolutional neural network architectures including DenseNet-121, VGG-16, MobileNet-V2, and ResNet-50 to identify the four most common rice leaves diseases in the Mekong Delta, Vietnam, such as bacterial leaf blight, tungro, blast, and brown spot, and obtained better performances compared to the original architectures with accuracies of 0.9930, 0.9703, 0.9740, and 0.9770, respectively.