Enhancing Apple Leaf Disease Classification Using Transfer Learning and CLAHE-Based Image Preprocessing

The diseases of the apple leaf are among the factors affecting agricultural production; thus, there is a need to detect them efficiently and accurately. The study investigates how CLAHE image preprocessing affects the cateGorizing of appLe leAf disEases using transfer learning with MobileNetV2. Plantvil- lage data containing four disease types were used to develop the classifier using raw (unprocessed) images and CLAHE enhanced images. An individual CNN model was also trained as the baseline model. Validation accuracy of 99.31%, 98.74% and 97.08% were achieved using raw image MobileNetV2, CLAHE enhanced MobileNetV2 and CNN, respectively. It is clear that CLAHE processing improves the local contrast of leaf images, making them more visually appealing. However, it did not improve the classification accuracy of the models. It implies that the feature representations of MobileNetV2 pretrained networks are good enough to extract the various features without necessarily im- proving the contrast. The model trained using CLAHE enhanced images still recorded an impressive macro-averaged F1 score of 0.98%. These results show that transfer learning works well for classifying plant diseases and give us more information about how image preprocessing works in deep learning pipelines.