This paper proposes an improved YOLOv8 model named A-YOLOv8s, which aims to improve the efficiency and accuracy of resistance spot welding (RSW) defect detection. The model is improved on the basis of YOLOv8s in the following ways: the Adown module is used to replace the original downsampling convolution in the Backbone part to enhance the multi-scale feature extraction capability; the C2f-GhostDynamicConv module is introduced in the Backbone and Neck parts to improve the efficiency of feature fusion; and the EfficientHead is used in the Head part to optimize the detection head structure and reduce the amount of calculation. Experimental results show that A-YOLOv8s outperforms YOLOv8n and YOLOv8s in terms of mAP50, and achieves a good balance in terms of parameter quantity and GFLOPs, which shows that the model can effectively improve the detection performance while maintaining lightweight. The improved model of this study provides an efficient and automated solution for RSW defect detection, and has the potential to be widely used in industrial automated detection.

The research and development in the field of robotic arm industry has brought opportunities for the industrial manufacturing of medical prostheses. Aiming at the difficulties and accuracy of knee joint prosthesis surface deviation polishing in traditional medicine, an improved digital detection, analysis and polishing method based on M3C2 algorithm is proposed in this paper. In this method, 3D scanning technology is used to measure prostheses, complete 3D point cloud data is obtained, and the obtained standard and non-standard model point clouds are preprocessed. The two point clouds were registered by the double registration module of Maximal cliques and ICP algorithm. The improved M3C2 algorithm combined with the local similarity algorithm was used to compare the point clouds and detect the error, and the shape error, error distance and direction of the surface points of the knee joint prosthesis were quantified. Experimental results show that the method is effective and feasible. The method proposed in this paper can effectively extract the quantified error location, error distance and direction, which provides a feasible scheme for the automatic grinding and polishing of knee prosthesis by robotic arm, and an effective simulation experiment is carried out.

In modern biological research, rodents play a crucial role as key experimental subjects in numerous fields. Accurately tracking the movements and behaviors of rodents is of great significance for obtaining valuable research data. Traditional tracking methods, such as marker - based techniques, often interfere with natural behaviors and fail in complex scenarios. This paper proposes a novel multi - object rodent tracking algorithm enhanced by re - identification. This algorithm combines motion features and re - identification features to reduce identity switches and false detections, enabling precise tracking of rodents. Experimental results on the AnimalTrack dataset show a significant improvement in the MOTA and IDF1 metrics, especially in complex scenarios. The proposed algorithm provides a robust solution for rodent behavior analysis and has potential applications in neuroscience, ethology, and drug research.