The aim of this study is to construct a deep learning-based biomechanical model of musical instrument playing action that integrates skeletal pose estimation and action recognition techniques. PHRNet-based human pose estimation can extract the skeletal key points of a player from video data, and these key points provide basic data for instrumental performance action recognition and analysis. The human skeletal action recognition method based on diversity rewarded reinforcement learning framework (DDRL-GCN) classifies the extracted key point sequences into specific playing actions, and the musical instrument playing actions are successfully modeled. The biomechanical model of musical instrument playing action designed in this paper is applied to recognize the playing action of five different musical instruments, and the recognition accuracy can reach more than 90%. This paper is designed to distinguish between different musical instruments, the recognition effect is more satisfactory.