Basketball teaching methods based on 3D-Convolutional neural network

To address the shortcomings of current basketball teaching methods, which rely heavily on human experience, have low teaching efficiency, and make personalized feedback difficult to achieve, the intelligent recognition of technical actions has become an important development direction in teaching informatics. However, existing recognition methods typically have limitations, such as inadequate modeling capabilities for complex actions, low accuracy under occlusion and perspective changes, and difficulty meeting real-time requirements. The study optimized the single-shot multibox detector algorithm by introducing the focal loss function and other means. The optimized algorithm is then used for video skilled movement detection processing to obtain cropped frame images. The basketball skilled movement dataset is classified and processed using a support vector machine, while a dual-resolution 3d-convolutional neural network is employed to process action features. The results revealed that the improved single-lens multi-box detector algorithm processed images faster and with higher accuracy and F value compared to the other algorithms. In the INRIA dataset, the improved single-shot multi-box detector algorithm processed 67 frames per second, which was 25 frames more than the single-shot multi-box detector algorithm. Its accuracy rate in the PASCAL VOC dataset was maximum 0.908, which was 0.105 greater than the pre-improved algorithm. Compared to algorithms such as 2d-convolutional neural network, the dual-resolution 3d-convolutional neural network had higher accuracy and smaller mean absolute error values. The F1 value of the combined model of this algorithm was 0.174, which was 0.102 smaller than that of the 2d-convolutional neural network. The classification accuracy of the combined model of this algorithm was larger at 0.925. Compared with traditional teaching methods, this approach quantitatively assesses students' technical levels and improves the relevance and scientific nature of teaching feedback.