Artificial Intelligence Assisted Early Blight and Late Blight Potato Disease Detection Using Convolutional Neural Networks

Abstract

Developing countries like Ethiopia have resources suitable for the production of different varieties of crops. Potato is the fourth as a major crop in  the world, after rice, wheat, and maize. In Ethiopia, one of the crops produced and consumed in mass is potatoes. Nonetheless, the yield per unit  area of potatoes is very low compared to other countries. There are a plethora of reasons and one of them is potato disease. The major disease,  which affects the major potato production area is late blight, according to researchers on the field estimated losses range from 6.5 to 67.7%  depending on the accessibility of varieties. As plant pathologists mentioned not to take early late blight disease management would destroy the  whole farm within a few days. For decades many researchers have experimented on plant disease detection and classification using computer vision  via different approaches and algorithms. Most researchers used traditional machine learning algorithms that require a handcrafted feature  extraction to classify and detect a given image as per its classes. The contribution of this work is twofold, using deep learning for potato disease  detection and developing an AI-based android application prototype. An Image dataset that is labeled with three classes as ‘Healthy’, ‘Early blight’,  and ‘Late blight’ is used as a benchmark. The pretrained models of deep learning, MobileNet, and EfficentNet have shown 98% prediction  performance. Finally, the model was built integrated with the android application and tested with unseen data.