Application of Improved YOLOv8n-seg in Crayfish Trunk Segmentation

Abstract

The crayfish industry (Procambarus clarkii) is experiencing rapid growth. However, the processing sector continues to face challenges due to a lack of advanced automation, relying heavily on manual visual inspection to assess crayfish specifications and integrity, which limits efficiency and precision in decision-making. To address the issue of intelligent grading of P. clarkii, this work proposes the GHB-YOLOv8-seg algorithm for segmenting the main trunk of P. clarkii shrimp based on the YOLOv8n-seg model. The original main trunk network is replaced through the coupling of Ghost and HGNetV2, and depth-separable convolution is employed to perform the linear transformation of the features. This results in a reduction in the number of parameters and computational complexity while maintaining high accuracy. The computational complexity is reduced; concurrently, introducing the weighted bidirectional feature pyramid network (BiFPN) enables the model to perform multi-scale feature fusion with greater alacrity, thereby enhancing the model's performance. Ultimately, the intelligent grading of crayfish specifications was achieved by calculating the pixel area after segmentation and converting it to the actual body weight. The results demonstrated that the number of parameters of the improved YOLOv8n-seg model was reduced by 60.5%, the model size was reduced by 55.4%, and the mAP value was increased from 98.9% to 99.2%. The study indicates that the YOLOv8n-seg algorithm model facilitates precise and lightweight segmentation of the crayfish trunk, which can be integrated into diverse mobile devices.