A Comprehensive Survey on Potato Leaf Disease Detection Using Machine Learning, Deep Learning, and Explainable Artificial Intelligence Techniques

Source abstract

Potato (Solanum tuberosum L.) is considered one of the most significant food crops across the globe. In addition, potatoes are highly valuable for ensuring food security. Nevertheless, potato cultivation is significantly hampered due to different kinds of diseases such as Early Blight, Late Blight, Bacterial Wilt, and Potato Virus Y (PVY). Traditional disease recognition technologies are based on the manual detection of potato diseases performed by agricultural experts. As a result, disease diagnosis requires much time and effort and is prone to errors. The development of innovative Artificial Intelligence (AI), Machine Learning (ML), Deep Learning (DL), and Computer Vision technologies has allowed the development of advanced algorithms that can be used for detecting plant diseases in an automated manner. The research provides an overview of main potato diseases, publicly available data sets, image pre-processing techniques, traditional machine learning algorithms, deep learning models, transfer learning techniques, and Explainable Artificial Intelligence (XAI). Moreover, the paper pays special attention to popular data sets such as PlantVillage, PlantDoc, Kaggle Potato Leaf Disease Data Set, and Potato Leaf Dataset (PLD). Besides, the review discusses state-of-the-art models including SVM, RF, CNN, VGG16, ResNet50, DenseNet121, EfficientNet, vision transformers, and hybrid deep learning techniques. The paper captures some of the recent trends in lightweight architectures, real-life disease detection, disease severity prediction, and explainability using Grad-CAM, SHAP, and LIME. A comparison of current work has been done to highlight the problems that exist, such as variation in the environment, a small number of datasets acquired from the field, class imbalance, computational complexity, and the model generalization problem. The paper concludes by looking into future trends such as edge computing, federated learning, multimodal sensing, transformers, and intelligent precision agriculture systems.