Potato Leaf Classification Using CNN

Abstract

The potatoes are among the most highly cultivated vegetables worldwide and elephantine cultivation of this crop forms a large part of the agricultural economy of India. But the production of potatoes is plagued by very high losses due to diseases like late blight and early blight. These losses incur yield losses and escalate production costs. The present investigation stresses an automated fast disease detection system that permits rapid diagnosis of diseases and works toward efficient potato production practices. The system detects and classifies disease images based on image processing techniques and deep learning algorithms. For this purpose, image features are extracted using a lightweight deep learning model-Mobilenet. Finally, these features are classified as healthy and diseased using an SVM classifier. There are more than 2000 images of healthy and diseased potato leaves that form the training and testing datasets, which have been downloaded freely from several online databases such as Kaggle. System Accuracy: 91.41% (with training and testing conducted at a ratio of 70-30). Results show that such a combination of a MobileNet feature extraction engine with SVM classification can improve the accuracy of detection, thus making this a feasible and flexible solution for automatic disease management in agriculture.