Automated Pathological Assessment of Potato Leaf Diseases through Convolutional Neural Networks

Abstract

Potatoes are grown all around the world at a large scale and are at the fourth number in the massive growth list. However, potatoes are primarily affected with fungus, resulting in early and late blight diseases, reducing the production rate of crops. Therefore, control and management of disease in real-time could help farmers enhance production, reduce crop, financial losses. Disease identification in plants is a potential step toward sustainability and security of the agriculture sector. Imaging-based processing, in particular, allows the in-depth study of plant physiology quantitatively. On the other hand, interpreting manually needs a lot of work, understanding of plant pathogens, and a long processing time. Therefore, this study proposes a time-efficient algorithm based on transfer learning and image processing that can accurately classify potato diseases. The proposed method consists of three steps preprocessing (grayscale conversion), segmentation (image enhancement, soft clustering, morphological dilation, and flood fill operation), and classification (AlexNet). This framework is tested over the three classes of the plant village dataset of the potato crop. Experimental results demonstrated satisfactory results and achieved 97.57% classification accuracy.