A Reliable Apple Leaf Disease Identification Using a Deep Learning-Based MobileNetV2 to Safeguard Apple Fruit Safety

Apples rank as the fourth most produced fruit globally. Nevertheless, substantial annual losses in apple yields are attributed to various leaf diseases. Early detection of these diseases is crucial for enhancing apple production. This research demonstrates the application of deep learning in accurately diagnosing apple leaf diseases from images without human intervention. We utilized the Apple Leaf Disease Dataset, which includes images of three categories of apple leaves: scab, rust, and healthy. The study introduces a customized five-layer head into the MobileNetV2 architecture. This modified version, termed TL-MobileNetV2, replaces the original classification layer with the new customized head. The TL-MobileNetV2 model achieved outstanding results, with an overall accuracy of 98.27%, a precision of 98.36%, a recall of 98.27%, and an F1 score of 98.28%. Additionally, Class Activation Maps were employed to generate heatmaps for disease detection, visually highlighting the regions that influenced the model's predictions.