Potato Disease Diagnosis with Symptoms Visualization

The devastating effects of global climate change on agricultural production have required regular monitoring of crops to detect possible diseases. Plant diseases are favored by the plants' stress, making them vulnerable to different environmental attacks. Herein, a new trend from artificial intelligence and computer vision emerges in crop monitoring; it involves tracking the health status of plants based on their morphological or physiological traits, the leaves being one of the plant constituents where diseases are mainly seen. In northern Africa, potatoes are a widely consumed food, and their harvest is a food security issue. This article addresses the detection and classification of potato diseases to allow early detection and treatment of diseases using a convolutional neural network based on potato leaf images. The deep learning-based models lack intrinsic explanation; this paper focuses particularly on the visualization aspect to address the results' explanation. The presented work addresses the explanation aspect of the decision by prospecting the feature maps in the convolutional layers and utilizing the Grad-CAM technique to visualize the detected symptoms that lead to the final decision through the classification layers.