Prediction of Grape Disease using Deep Learning based on MobileNetV2 and Explainable AI with Grad-CAM for Disease Management

The grapes production can be affected by diseases that cause changes to the look and state of the leaves and may reduce the quality of the crop if the disease are not detected in time. Even though visual diagnosis is of some help, it depends on the operator's experience and can become difficult when different diseases show similar symptoms. The study presents an image-based approach for predicting and Gradient-weighted Class Activation Mapping (Grad-CAM) for providing a visual explanation. The System can detect four kinds of disease: Black Measles, Black Rot, Healthy, and Isariopsis Leaf Spot. The model produces a distribution of class probabilities and Grad-CAM is used to highlight the parts of the image that are related to the predicted class. A general module based on knowledge then provides information regarding the symptoms, advice on how to manage the disease, and preventive measures for the disease which has been predicted. Having been tested on 903 images, the system achieved 886 correct predictions and 17 incorrect ones, giving an accuracy of 98.12%, a weighted precision of 98.13%,a weighted recall of 98.12%, and a weighted F1-score of 98.12%. The errors that were seen were mostly between Black Measles and Black Rot. The study therefore incorporates prediction, visual explanation, and disease-management information into one workflow, while at the same time recognizing the need for further validation with a wider variety of field images..