Deep Learning-Based Attention Models for Behavior Recognition in Volleyball Video Analysis

With the Ministry of Education’s vigorous promotion of curriculum reform, the integration of “[Formula: see text]” has rapidly advanced. In this era of technological progress, the convergence of online and offline teaching models is inevitable, relying heavily on the synergy between network information technology and higher education methodologies. This integration not only enriches traditional teaching but also revitalizes physical education in colleges. Beyond conventional offline training, public sports volleyball instruction at universities should encourage students to engage with video behavior recognition technologies. As electronic equipment and technology evolve, the pressing need for enhanced video understanding drives researchers to focus on various aspects of video analysis. In this study, we utilize computer deep learning techniques to explore and analyze spatial attention, channel attention and dual attention within the open volleyball video dataset named Volleyball. Our experimental findings validate the efficacy of our algorithms, demonstrating superior performance compared to state-of-the-art methods. The results indicate a high accuracy rate for group recognition, reaching 87.8%, which is compelling evidence of our approach’s effectiveness. Additionally, using the same research data and methodology, our algorithm achieved an 83.8% accuracy rate for individual recognition, with a single instance achieving 81.8% accuracy. These outcomes underscore the potential of integrating advanced technologies into educational practices, particularly in enhancing the teaching and learning experience in physical education.