Paddy Leaf Disease Detection Employing Transfer Learning

Approximately 135 million people in Bangladesh use rice as their primary source of food. Nowadays, paddy diseases constitute a serious agricultural problem that influences both rice production and food security as well as rural livelihoods. However, owing to the prevalence of several diseases that affect crops, the sustainable production of rice is a major hurdle. Losses in yield are caused by prevalent diseases such as blast, sheath blight, and bacterial leaf blight. There are roughly 30 paddy leaf diseases in Bangladesh, and 7-8 of them are fairly prevalent. These diseases cause significant economic losses for both farmers and the national economy as they not only lower crop production but also decrease grain quality. To address these issues; researchers in Bangladesh; have been working hard to understand the biology of paddy diseases and to create integrated disease control systems. However, physical detection of diseases affecting rice leaves requires skilled professionals and time. The harm to crops can be significantly reduced, and farmers’ losses can be avoided if these diseases are quickly and accurately recognized at an early stage. In this study, we have proposed a model based on convolutional neural networks (CNN) that can identify four categories of paddy diseases as well as healthy leaves. The primary objective in this research is to offer the best results for paddy leaf disease identification using deep CNN models. In contrast to the long, traditional manual medical diagnosis approach, where accuracy is also highly debatable, the Yolo-V8 can attain the maximum level of accuracy. The Yolo-V8 model evaluated four models: Yolo-V8, Inception-Resnet-V2, ResNet-101, and Xception, and Yolo-V8 achieving 99.16% accuracy.