A Novel Deep Learning Method Based on Modified Recurrent Neural Network for Sports Posture Recognition

Abstract

ABSTRACT In the process of studying the sports posture recognition algorithm, due to the influence of the shadow noise generated in the movement, when the current algorithms are used for the sports posture recognition, the extracted sports posture features have limitations. And there is a problem that it cannot fully reflect the dynamic characteristics of the posture. To solve the above problems, through the deep transfer learning experience in image recognition, we propose a novel deep end-to-end neural network model based on the Inception neural network and recurrent neural network. The new model adopts the 1×1 convolution to combine the multi-channel data. The convolution of different scales extracts the waveform characteristics of different scales. Maximum pooling filters false positives caused by minor perturbations. Time feature extraction is used to model time series features. It makes full use of data characteristics to complete the classification task. Compared with the state-of-the-art neural network models, the results show that sports posture recognition accuracy is improved by nearly 3 %.