Objective This study aims to address the problems of single methods, low efficiency, and high costs of the quality inspection of crayfish in industrial processing.Methods A YOLO-HDR-based lightweight neural network model was proposed. The PP-HGNetv2 model was employed to design a new YOLOv8 backbone network, and the lightweight modules of HGstem and DWConv were introduced to reconstruct the network. The dynamic convolution block and other lightweight convolutions (GhostConv and RepConv) in the official library were used to redesign the HGBlock of the new backbone network. The dynamic high-performance network modules (DynamicHGBlock, RepHGBlock, and GhostHGBlock) were obtained to improve HGBlock and the feature expression of the network. The C2f module of the original neck network was improved by the repeated cross-stage local edge-preserving attention network RepNCSPELAN4 to address the performance degradation caused by the lightweight network.Results The accuracy and average precision of the improved model reached 92.8% and 95.9%, respectively, which were 3.5% and 1.9% higher than those of the original model and better than those of other comparative target detection algorithms. The number of parameters and model size of the improved model were reduced by 17.7% and 16.2%, respectively, compared with those of the original YOLOv8n model, and the amount of computation was reduced by 19.8%.Conclusion The method established in this study demonstrates improved detection performance under the dense occlusion noise background, enabling the quality inspection of crayfish in industrial processing in the complex background before frozen packaging.