Edge computing-enabled lightweight cross-modality attention fusion for IoT-based physical education systems

Emerging technologies such as the Internet of Things (IoT), edge computing, and virtual reality (VR) have brought significant changes to educational methods, particularly in physical education. However, traditional approaches still face challenges in providing accurate performance assessments and real-time data-driven decision support, especially in complex and large-scale environments. To address these issues, we propose STC-EdgeNet, a lightweight multimodal fusion model that combines spatio-temporal graph convolutional networks (ST-GCN), TinyLSTM, and a cross-modal gating mechanism for real-time action recognition and quantitative performance evaluation. The model is optimized for edge computing, ensuring low latency and energy efficiency, which are critical for real-time data processing in resource-constrained environments. STC-EdgeNet improves action recognition accuracy while enabling reliable performance assessment via multimodal data fusion. Experimental results show that STC-EdgeNet outperforms state-of-the-art models, achieving 97.6% accuracy on NTU RGB+D 60 (X-View) and 92.1% on NTU RGB+D 120 (X-Sub), with high computational efficiency. This work demonstrates that STC-EdgeNet can provide fine-grained performance evaluation and real-time data support, offering a powerful solution for intelligent physical education systems based on edge computing and multimodal recognition.