Convolutional neural network-based feature extraction and quantification of daily sports data for smart wearable devices

With the progress of modern technology, smart wearable devices have been gradually applied in the field of sports.This paper focuses on the experiments of motion recognition of the main joints realized by convolutional neural network-assisted smart wearable devices.Using smart wearable devices to feature extraction of a variety of sports signals, using GAF algorithm for sports signal image coding, and using convolutional neural network and gated recurrent unit, a CNN-GRU-based motion recognition method is proposed.Through the training and evaluation experiments of the model, it is found that the average accuracy of the CNN-GRU model training and testing is higher than 96%, and the loss value is lower than 1.5%, and the performance of sports recognition is better than that of CNN and CNN-LSTM models.Meanwhile, it presents excellent performance in the recognition of sports with different classifications and different signal durations, reaching 97.02% and 92.63% accuracy in the recognition of three and four types of sports, respectively, and the distribution of the values of human body indexes in different sports in the case study presents a certain degree of regularity, which verifies the effectiveness and feasibility of the CNN-GRU model in different application scenarios.It also shows that the method has great development potential in the field of intelligent sports.