Addressing the challenges of localizing true lesion areas in fish disease detection due to irregular lesion shapes, unclear textures, and scattered lesion spots, a CEH-YOLOv8 fish disease detection method based on dual-channel layered collaborative is proposed. Firstly, a dual-channel feature extraction network is introduced to enhance the model's ability to extract irregular lesion areas with unclear textures. Subsequently, an Efficient Channel Spatial Attention (ECSA) mechanism is proposed to improve the model's recognition capability for distributed targets. Additionally, to reinforce the improved backbone network, a Hierarchical and Balanced Feature Pyramid Network (HBFPN) is presented, which integrates features extracted from the backbone network at different levels to enhance the model's feature representation capability. Experimental results demonstrate that the CEH-YOLOv8 network achieves an accuracy rate of 83.2%, a recall rate of 72.5%, and a mean Average Precision (mAP) of 76.2% for fish disease detection. Compared to the state-of-the-art YOLOv10 method, it exhibits improvements of 6.9, 11.6, and 11.9 percentage points, respectively, and enhancements of 4.3, 6.9, and 6 percentage points compared to the original model. The inference time for a single frame is 9.1 ms. These results indicate that the improved YOLOv8 network can accurately screen fish with diseases, enabling early detection of fishery diseases to reduce economic losses.