Evaluation of XAI Models for Interpretation of Deep Learning Techniques’ Results in Automated Plant Disease Diagnosis

Automated disease diagnosis in plants is crucial for the agriculture industry to maintain crop health and increase yields, which has significant implications for global food security and the economy. The use of convolutional neural networks (CNNs) for disease diagnosis has gained much attention due to their ability to detect patterns in the images of plant leaves, allowing for the accurate diagnosis of diseases. However, one of the major challenges in using CNNs is their limited interpretability, which makes it difficult to understand the reasoning behind the model’s output. In this work, we explore the potential of CNNs for plant disease diagnosis and propose the use of explainable artificial intelligence (XAI) methods to improve the interpretability of the CNN’s output. We first discuss the state of the art in plant disease detection, including conventional methods such as Polymerase Chain Reaction (PCR) and Isothermal Amplification based diagnosis, and the rise of CNNs in the field. We then introduce the architecture and the principles of CNNs, highlighting their ability to classify images and their use in plant disease diagnosis. However, due to their black-box nature, the interpretation of CNN outputs remains challenging. To address this, we propose the use of post-hoc XAI methods, specifically LIME, SHAP, and Grad-CAM, to provide insights into CNN’s decision-making process. Our study aims to demonstrate the potential of CNNs for plant disease diagnosis and the importance of interpretability in deep learning models. We hope our work will contribute to the development of more accurate and interpretable CNN models for disease diagnosis in plants, ultimately leading to more efficient and sustainable agriculture practices.