A Dual-Branch Model Integrating CNN and Swin Transformer for Efficient Apple Leaf Disease Classification

Abstract

Apples, as the fourth-largest globally produced fruit, play a crucial role in modern agriculture. However, accurately identifying apple diseases remains a significant challenge as failure in this regard leads to economic losses and poses threats to food safety. With the rapid development of artificial intelligence, advanced deep learning methods such as convolutional neural networks (CNNs) and Transformer-based technologies have made notable achievements in the agricultural field. In this study, we propose a dual-branch model named DBCoST, integrating CNN and Swin Transformer. CNNs focus on extracting local information, while Transformers are known for their ability to capture global information. The model aims to fully leverage the advantages of both in extracting local and global information. Additionally, we introduce the feature fusion module (FFM), which comprises a residual module and an enhanced Squeeze-and-Excitation (SE) attention mechanism, for more effective fusion and retention of both local and global information. In the natural environment, there are various sources of noise, such as the overlapping of apple branches and leaves, as well as the presence of fruits, which increase the complexity of accurately identifying diseases on apple leaves. This unique challenge provides a robust experimental foundation for validating the performance of our model. We comprehensively evaluate our model by conducting comparative experiments with other classification models under identical conditions. The experimental results demonstrate that our model outperforms other models across various metrics, including accuracy, recall, precision, and F1 score, achieving values of 97.32%, 97.33%, 97.40%, and 97.36%, respectively. Furthermore, detailed comparisons of our model’s accuracy across different diseases reveal accuracy rates exceeding 96% for each disease. In summary, our model performs better overall, achieving balanced accuracy across different apple leaf diseases.