A hybrid SE-YOLOv5 and GNN-SGCM approach for intelligent table tennis ball positioning and 3D recognition

At present, all the positioning methods for the trajectory of table tennis have limitations such as low accuracy and large deviation.Therefore, this study utilises the extrusion and excitation network to optimise YOLOv5, introduces the graph convolutional network to improve the hybrid algorithm of semi-global matching and census transformation, and combines the two to construct an intelligent positioning and recognition model for table tennis.The results show that the research model has an accuracy rate of 97.6%, a precision rate of 98.6%, a recall rate of 96.8%, and a specificity of 97.2%.The average error of the recall rate is 0.59%, and the overlap degree of trajectory positioning is 0.93.In conclusion, the research model not only ensures the reliability of the table tennis positioning and recognition results, but also improves the recognition efficiency and result quality, making significant contributions to the development of table tennis.