DiRui Tan
School of Education, City University of Macau, Taipa, Macau 999078, China
Liyuan Huang
School of Digital Creativity, Guangdong Polytechnic Normal University, Guang Zhou, Guang Dong, 510665, China

Abstract:

In view of the problem that sports injuries caused by excessive exercise and poor posture in dance training cannot be corrected and fed back in time, this paper combines multimodal deep neural network (MDNN) and proximal policy optimization (PPO) model to explore the optimization of movement performance and physiological feedback in dance training through biomechanics and biosensing technology, so as to reduce sports injuries. The multimodal biological data of dancers is collected through motion capture technology, electromyography (EMG), electroencephalogram (EEG), and other body perception technologies, and MDNN is used to integrate multimodal information to provide dancers with a comprehensive analysis of movement execution and physiological state. On the basis of multimodal data integration, the PPO reinforcement learning model is applied to realize personalized dance training movement optimization according to the dancer’s physiological response and movement feedback, so as to avoid sports injuries. The results of experimental analysis show that the personalized dance movement training optimization method studied in this paper is superior to traditional training methods and training methods relying on biosensors, and obtains higher accuracy (0.91), fluency (0.89), coordination (0.93), stability (0.94), and flexibility (0.91). The sports injuries after optimized dance training are analyzed, and the frequency of muscle strain, joint injury, sprain, and soft tissue contusion during the warm-up period before and after training, the training period, and the recovery period after training is significantly reduced. This study provides an innovative technical solution for the field of dance training, especially in reducing sports injuries and improving training effects and personalized training.