ECA-ViT: Leveraging ECA and Vision Transformer for Crop Leaves Diseases Identification in Cultivation Environments

Crop disease identification is crucial for its role of maintaining crop yield and quality in precision agriculture. Traditional disease detection methods are often inefficient and error-prone. To address this issue, we propose the ECA-ViT model, which combines Efficient Channel Attention (ECA) and Vision Transformers (ViT) to identify crop leaf diseases in cultivation environments. The ECA-ViT model uses adaptive 1D convolution kernels for rapid cross-channel interaction of local disease features in crop images, compensating for the Vision Transformer's lack of global context understanding and optimizing the model. This balanced approach enhances feature learning while maintaining computational efficiency without increasing model complexity. We evaluated the ECA-ViT model using rice leaf disease datasets from real field settings. The results demonstrated that ECA-ViT outperformed existing deep lear-ning methods, achieving an accuracy of 95.41%.