Research on cutting keypoint detection of chicken carcasses based on improved YOLO11-Pose

Research on automated poultry processing systems is crucial to improve the productivity and ensure product consistency of industrial-scale production. In this study, a method for predicting chicken carcass cutting points, based on an improved YOLO11-Pose, was proposed to minimize segmentation errors and obtain precise cutting coordinates for guiding mechanical cutting devices, accommodating the size variations across different chickens. Initially, the SimAM attention mechanism was incorporated to improve the extraction of critical features from chicken carcass images without introducing additional parameters, thereby enriching feature representation for accurate cutting-point prediction. Subsequently, the C2PSA_BSAM channel-spatial attention synergy module was employed to perform dual-dimensional feature extraction, enhancing the ability of model to capture detailed features across diverse carcass conditions and morphological variations. Furthermore, the SPPF_Global module was further integrated to strengthen cutting-point detection by expanding the receptive field through multi-scale pooling, enabling more effective capture of global contextual information. Experimental results demonstrated that the proposed method outperformed the baseline model, with improvements of 6.6% and 5.2% in mAP50-pose and mAP50-95-pose, achieving values of 99.5% and 83.7%, respectively, for key anatomical locations. The accuracy of the visual system in identifying critical cutting points was further validated by the box_loss and pose_loss values, recorded at 0.27057 and 0.07895. Thus, the proposed methodology could provide precise technical support for automated chicken carcass cutting, enabling accurate, contact-free coordinate acquisition for key cutting locations across chickens of varying sizes.