Advanced Detection of Bean Leaf Disease with ResNet50: A Cutting-Edge Method for Precise Classification

The classification of bean leaf diseases is considered an important task in order to ensure effective management of the crop, thereby improving productivity. In this work, the authors have used the ResNet50 architecture for deep learning with respect to the classification of bean leaf diseases into three classes called Angular Leaf Spot, Bean Rust, and Healthy. Originally captured by the Makerere AI Lab in collaboration with the NaCRRI (National Crops Resources Research Institute) of Uganda in bean fields, this dataset was pre-split into training and testing and validation sets. However, the images were manually merged from these splits to have a single dataset for training the model. It achieved an overall accuracy of 88% with class-wise precision, recall, and F1-score values of 83.93%, 82.00%, and 100% for Angular Leaf Spot, Bean Rust, and Healthy categories, respectively. These reflect a highly effective performance, especially for the Healthy category, whose precision attained a perfect score. Indeed, the macro and weighted averages of precision, recall, and F1-score were 88.64%, 88.00%, and 88.10%, respectively. These findings accordingly illustrate that the classification performance through the ResN et50 model on bean leaf diseases is quite reliable and hence can be used in an agricultural setting for automatic disease detection.