An Explainable Deep Learning Approach for Potato Leaf Disease Detection and Management

Deep learning has significantly advanced plant disease detection using leaf images, offering high accuracy in classification tasks. However, these models often function as black boxes, lacking interpretability, a crucial aspect of agricultural decision-making. To overcome this challenge, this work introduces a deep learning based Explainable AI (XAI) method for a potato leaf disease detection system. The proposed technique uses targeted perturbations guided by intermediate object detection outputs, offering a more focused and interpretable visualization of regions influencing the model's predictions. The model was trained and evaluated on the PlantDoc real-world agricultural image dataset. Qualitative assessments confirm that the proposed explanations align more with human expert understanding, successfully highlighting disease-infected regions while ignoring irrelevant background noise. This work ensures high performance in disease classification and provides meaningful visual explanations, enhancing trust and usability in practical agricultural automation. The simulation results indicates that the proposed approach achieves a classification accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6. 8 \%}$</tex>, outperforming several existing approaches in precision and recall. Additionally, the saliency maps generated by our method demonstrated higher localization accuracy, with a 23 % improvement in Intersection-over-Union (IoU) scores compared to traditional gradient-based XAI methods.