Exploring Deep Learning Architectures for Multiple Apple Leaf Disease Classification

This paper delves into the selection and adaptation of deep learning architectures for classifying diseases in apple trees, with a particular focus on three widely recognized CNN models: VGG16, ResNet152V2, and DenseNet121. By harnessing the established performance and adaptability of these networks, we utilize transfer learning to initiate training with pre-trained models. This approach allows us to capitalize on learned feature representations and mitigate the need for extensive datasets. Our experimental results, conducted on both the PlantVillage and Turkey Plant datasets, showcase the efficacy of ResNet and DenseNet architectures, potentially attributed to their superior feature extraction capabilities. We outline the preprocessing steps, which include employing data augmentation techniques to augment diversity and enrich the training data. Importantly, our study demonstrates improved classification accuracy, highlighting the significance of architectural selection and customization in optimizing deep learning models for specific tasks and providing valuable insights for practitioners aiming to deploy efficient and effective classification systems.