Intelligent Volleyball action recognition integrating attention mechanism and multi-scale spatiotemporal separation algorithm

In intelligent sports analysis, accurate volleyball action recognition is a challenging task. Existing technologies suffer from low accuracy and insufficient robustness. Therefore, the research aims to achieve efficient data processing and feature extraction by integrating attention mechanisms with multi-scale separation spatiotemporal algorithms to optimize the accuracy and robustness of volleyball action recognition. The extensive testing is conducted on the UCF101 dataset and the custom volleyball technique action video dataset. The proposed algorithm demonstrated excellent performance, with recognition accuracy of 99.5% and 99.6%, respectively, significantly better than that of existing basic and advanced algorithms. In addition, this algorithm outperformed existing algorithms in terms of recognition time, computational complexity, generalization ability, and real-time performance, with normalized metrics of 0.97, 0.93, 0.95, and 0.94, respectively. These key data results fully present the effectiveness and superiority, which helps coaches and athletes better understand technical movements and optimize training plans.