AI for Agriculture: Detecting Black Gram Diseases with Deep Learning and Explainable AI

Black Gram is one of the most important crops in South Asia due to its high resistance to adverse climatic conditions and its ability to improve soil fertility. However, despite its resilience, the production of this plant suffers losses every year due to various diseases such as: Cercospora leaf spot disease, Anthracnose, leaf crinkle, Powdery mildew, Yellow mosaic, and insect infestation. Traditionally, the detection of these diseases has relied on manual assessment by agricultural experts, a time-consuming and labor-intensive process. Advances in artificial intelligence offer a promising solution by enabling early diagnosis of leaf diseases increasing efficiency and accuracy. This paper presents a transfer learning model for the classification of different leaf diseases in Black Gram that achieves high performance. The best performing model, DenseNet121, achieves an accuracy of 98.47%. Explainable AI (XAI) techniques were integrated to further improve the reliability and interpretability of the results. In particular, the Gradient-weighted Class Activation Mapping (Grad-CAM) method was used to visualize the regions of the leaf images that the model focuses on during disease classification, enabling a transparent and interpretable decision process.