The author in order to realize the skill assessment of soccer players and analyze the athletes’ performance on the field. Based on the extraction of multimodal data of soccer players and the fusion of multistream data with adaptive multiscale differential graph convolutional network, the athlete action recognition model is constructed. Next, the PC-CNN skill assessment model is constructed to assess and analyze the skills of soccer players. The performance performance of the assessment model in this paper is tested through comparative experiments. Finally, analyze the performance of the soccer team in the shooting mode and shooting area. The accuracy of this paper’s PC-CNN assessment model is more than 90% in the assessment of three levels of skills of soccer players, and the accuracy of the assessment of high-level skills reaches 100%, which is more accurate than other models. Among the multimodal data of athletes’ physiology, joint movements had the greatest influence on the assessment accuracy of the assessment model, which reached 78.17%.The soccer teams A, B, and C had the highest success rate in the item of penalty kick shooting skill, with success rates of 75%, 66.7%, and 66.7%, respectively. Teams A and C had the highest success rate in Zone 1.Team B had the highest success rate in Zone 4.
