Basketball Target Recognition Based on Deep Learning

In order to study the analysis of athletes’ pictures in basketball game and the problem of action foul or not in the process of sports. In this paper, the deep convolution neural network method is used for target recognition, the athletes’ action posture and behavior data are extracted, and the complex technical action behavior in basketball is tracked. Combined with the characteristics of CNN image extraction and action recognition algorithm, the model is constructed. The following results are obtained: In the feature point calibration experiment, the error rate is not higher than 0.4%, and the average is 0.185%, which shows that the calibration method in this paper has high accuracy; In the process of denoising, the PSNR of the method proposed in this paper is better than GSM, KSVD, and CN2 methods, slightly better than multi-layer perceptron networks (MLPs), and the time consumption of processing the same picture is much lower than other methods. The accuracy rate of this method is the highest among all algorithms, reaching 95.6%, which shows that the target detection system designed in this paper is effective. The application of the basketball technical action recognition model promotes the training of difficult basketball technical actions, avoids sports injuries, and provides theoretical support for better improving the movement skills of basketball players.