A Multi-Source Data Visual Recognition and Grading Model for Student Pickleball Skill Levels in University Physical Education

Accurately assessing students' motor skill levels is crucial for implementing differentiated instruction and precision teaching in university physical education (PE) courses. Addressing the significant variation in student skill levels within pickleball PE classes, this paper proposes a visual representation learning-based model for predicting student performance. This research integrates multi-source teaching data to construct eight feature dimensions, including gender, general sports foundation, pickleball training duration, serve success rate, technical stability, match point differential, physical fitness test index, and inclass performance evaluation. The core innovation of the model lies in its transformation of traditional one-dimensional feature data into image-like two-dimensional heatmaps, which are subsequently processed using an improved YOLOv8 classification model. To enhance the network's feature extraction capability for this specific data pattern, a Lightweight Adaptive Extraction (LAE) module is introduced into the backbone network. Experimental results demonstrate that our model achieves a top accuracy of 96.3% on a real-world teaching dataset, outperforming traditional methods. This solution proves to be an efficient and precise tool for grading student pickleball skill levels, providing reliable decision support for PE teachers in implementing differentiated instruction.