Figures and Tables

A novel Wushu Sanda movement recognition method is proposed using semantic feature matching of significant images. By analyzing human skeleton distances and angles, spatial motion features are captured to generate motion feature maps. Singular value decomposition compresses data, removing redundancy and enhancing key pixel importance. An improved deep CNN is then employed for recognition. Experimental results show high accuracy, with a maximum Gini index of 0.10, stable recognition loss around 0.1, and an F1 score close to 1, proving the method’s feasibility.