Track and field sports skill recognition based on multimodal sensing data

Track and field sports skill recognition is a key technology in intelligent sports training, but traditional methods suffer from issues such as information redundancy and poor recognition performance.To address this, this paper first proposes an adaptive selection mechanism for multimodal sensor data based on mutual information, filtering out sensor combinations that provide maximum information correlation.Then, a convolutional neural network (CNN) is combined with a long short-term memory network (LSTM) for multimodal sensor feature extraction, and a recurrent matrix-based multimodal feature fusion method is proposed.Finally, the fused feature vector is input into a fully connected layer, and the softmax function is used to calculate the score for each category of athletics skill from the output classification layer.The experimental results show that the Macro_F1 of the proposed method is improved by at least 4.01% compared to baseline methods, demonstrating good recognition performance.