A Deep Transfer Learning-Based Approach to Detect Potato Leaf Disease at an Earlier Stage

Abstract

Bangladesh is an agricultural country and potato is the most cultivated crop here and even worldwide. But the production of potatoes is declining day by day due to various potato leaf diseases which can result in significant environmental and economic damage. It's difficult for farmers to find out which diseases damaging crops because they tend to use the traditional approach and the result is not accurate always. For that, it’s hard to take a decision on which fertilizers to apply. This traditional approach is a more time-consuming and slow process. To detect leaf diseases of potato at the early stage, this study present a deep learning-based approach using ResNet50. Using this technique, farmers can find out the actual diseases of potato in a feasible, efficient and time-saving way at their early stage and able to take fast decisions. That will help to grow more potatoes. It can be ensured that this model can bring many benefits in the agricultural field both economic and ecologic sides. This study works on the most two common diseases of potato leaves including late blight, early blight, and one healthy leaf. To find out the best model, this study has chosen 3 neural networks. After analyzing CNN, VGG19, and ResNet50 models get the accuracy according to 84%, 93%, and 97% for a collection of 2,152 images. In this paper, ResNet50 model achieves the highest accuracy.