부산대학교 도서관

As an essential food crop, rice is often infested with diseases that can cause significant yield losses and seriously damage economic income and food health. Early identification and control of rice diseases is an effective way to alleviate these problems. However, manual identification and diagnosis of rice leaf diseases requires experienced specialists and is time-consuming. In our study, we propose the ECA-ConvNeXt model, based on the ConvNeXt network, which can identify six categories of typical rice leaf diseases and healthy rice leaves. We also established a rice leaf disease identification dataset that contains images of healthy and diseased rice leaves with complex backgrounds and their disease category labels. In the proposed ECA-ConvNeXt model, we incorporated the ECA (Efficient Channel Attention) module, which improved the feature extraction performance using only a few parameters. Transfer learning was applied to load pre-training weights and fine-tuning was used to reduce training costs and improve the model performance. We tested the performance of ECA-ConvNeXt on the rice leaf disease identification dataset. Experimental results show that the proposed model achieved an accuracy of 94.82%, a precision of 94.47%, a recall rate of 94.31%, and an F1-Score of 94.33% on the rice leaf disease identification dataset. These results suggest that the proposed network effectively identifies rice leaf diseases.