Grape Leaf Disease Recognition Using ConvNeXt with Attention-Enhanced Feature Networks

This paper presents AF-ConvNeXt, an attention-enhanced feature network for grape leaf disease recognition. The proposed architecture integrates Convolutional Block Attention Module and Feature Pyramid Network into the ConvNeXt backbone to improve feature representation and multi-scale feature fusion capabilities. The CBAM is strategically placed at multiple locations to enhance both channel and spatial attention, while the FPN structure enables effective multi-scale feature extraction and fusion. We evaluate our model on a comprehensive grape leaf disease dataset containing four classes. Experimental results demonstrate that AF-ConvNeXt achieves 99.93% accuracy on the test set, outperforming conventional deep learning architectures. The model exhibits lower training loss compared to baseline models, validating the effectiveness of our proposed architectural improvements. These results suggest that AF-ConvNeXt offers a reliable solution for automated grape leaf disease recognition.