Potato leaf disease detection through ensemble average deep learning model and classifying the disease severity

Abstract

The varying crop species, symptoms of crop diseases, and environmental conditions make early detection of potato leaf disease difficult. Potato leaf diseases are difficult to identify in their early stages because of these reasons. An ensemble model is developed using the ResNet50V2 and DenseNet201 transfer learning algorithms in this study for identifying potato leaf diseases. For this work, 5,702 images were collected from the potato leaf disease dataset and the Plant Village Potato dataset. The datasets include valid, test, and train subdirectories, and the images are taken on 5 epochs. By including three more dense layers in each model and then ensemble that model, the performance of leaf classification may also be improved. Accurately and appropriately, the suggested ensemble averaging model identifies potato leaf phases. So, the accuracy of the suggested ensemble model is achieved with perfect precision. On the second level, the severity of the disorder is assessed using the K mean clustering algorithm. To determine the disease's severity, this system examines each pixel in the early and late blight images. It will be classified as severe if more than 50% of the pixels are damaged.