Intelligent Sensor Fusion and LSTM-Transformer-Based Model for Sports Behavior Prediction in Teaching and Training

In sports teaching and training, intelligent sensor fusion technology provides comprehensive and real-time data support for athletes through high-precision sports data acquisition and efficient fusion processing capabilities. Subsequently, this paper describes the process of smart sensor data fusion in detail, including data preprocessing (such as data cleaning, filling missing values, correcting errors), feature extraction, data association and data fusion, to ensure the quality and consistency of the subsequent input motion information data. In the prediction of sports behavior, the traditional LSTM has high computational complexity and resource consumption when dealing with long sequence data. To solve this problem, this paper proposes an algorithm based on LSTM and Transformer, which can capture the long-term dependence in time series motion data and improve the efficiency and quality of gesture sequence extraction. In order to further improve the prediction accuracy, the discrete cosine transform (DCT) technology is introduced to remove the miscellaneous information and only retain the main features. In the experimental simulation, the MPJPE results of different algorithms in walking, jumping, gymnastics and other movements are compared to verify the effectiveness of the proposed algorithm. The experimental results show that the MPJPE results of LSTM [Formula: see text] Transformer algorithm are better than those of other algorithms in the prediction time, which solves the problems of accuracy and efficiency in sports behavior prediction and action analysis in sports teaching and training.