A study and comparison of deep learning based potato leaf disease detection and classification techniques using explainable AI

The goal of artificial intelligence (AI), a field with a solid scientific foundation, is to enable machines to simulate human intelligence and problem-solving abilities. AI focuses on the study, development, and application of complex algorithms and computational models, with a particular focus on deep learning techniques. The application of artificial intelligence to potato leaf disease detection can reduce the restrictions brought on by the artificial selection of spotted disease features and improve the efficiency and speedup. It has also turned into a research hotspot in the agricultural sector. This work consists of four types of potato leaf diseases, such as early-blight disease, septoria disease, late-blight disease, and black-leg disease. It is a challenging task to identify and classify such diseases from the healthy images. As a result, this work uses a set of benchmark deep learning models to identify and categorize these four disease types in potato leaves. Furthermore, compared to existing models, our recommended models provide better accuracy and have visible results. In comparison to other cutting-edge models, the results of the proposed deep ensemble algorithm (CNN, CNN-SVM, and DNN) offers the best accuracy of 99.98%. All the sample images (healthy and unhealthy) are collected from different farms of the West Bengal state and prepare the experimented dataset. The working model has an additional benefit in terms of running time complexity (O(Ei)(1ik) and 17.86 s) and statistical comparison. Finally, LIME and SHAP are used to evaluate the findings, create more trust, and improve performance by providing explanations for predictions.