Basketball Posture Recognition Using Neural Networks

Abstract

Recognizing and training basketball athletes on their postures is crucial for enhancing performance, preventing injuries, and optimizing movement efficiency on the court. Therefore, this paper employs a convolutional neural network (CNN) to recognize six training postures in basketball. In terms of model structure, four convolutional layers are designed to extract critical features for identifying the six postures. To maintain consistency between the extracted features and the original features, this work uses the optimal mass transport (OMT) map to derive the model's loss function. Finally, the proposed model is evaluated on image datasets. Experimental results demonstrate that the proposed model outperforms competing methods in recognizing the six training postures. We find that the loss function derived from the optimal mass transport map significantly improves the CNN's image recognition capabilities. This is because the OMT map preserves the geometric characteristics of the original data distribution to the greatest extent possible.