FCNN-ItemCF Model for Personalized Training Strategy Recommendations in College Football Courses

With the increasing popularity of football, comprehensively improving students’ physical health has become an important direction. The concept of personalized education is gradually emerging. Especially in physical education, how to conduct customized training according to students’ specific needs has become a hot research topic. Traditional sports education design adopts a unified teaching method, ignoring the individual differences of students, resulting in students being unable to conduct targeted training, which affects the learning effect. In football data processing, traditional Convolutional Neural Networks (CNN) and other models have large parameters and low training efficiency. Taking college football courses as an example, this paper combined the advantages of Fully Convolutional Neural Network’s (FCNN) full convolutional layer and end-to-end training, as well as the Item Collaboration Filter (ItemCF) algorithm, to conduct personalized recommendation research on training strategy content in sports education design. The study first constructed an FCNN, used it to extract key action, posture, speed and stride feature maps from video frames, and combined it with the global average pooling layer to output the feature map as a comprehensive feature vector. Then, the user feature vector and user historical behavior data were fused and input into the ItemCF algorithm to calculate the project similarity and recommendation weight and generate a personalized training project recommendation list. Finally, a college football course education design platform was built, and the recommendation results were provided to students, who were then trained according to the recommended items. The experiment was based on the athlete training video data of college football courses in the public dataset Roboflow and the self-collected FB video dataset, in which user historical data collected the evaluation data of the experimental group and the control group in key action skills such as passing, ball control, shooting and physical endurance, speed and agility. Finally, a questionnaire survey was conducted to collect students’ satisfaction with the college football course recommendation system designed. The experimental results show that the recommendation accuracy of the Fully Convolutional Neural Network-Item Collaboration Filter (FCNN-ItemCF) model reached 92.35%, which is 6.7% higher than that of the Convolutional Neural Networks-Item Collaboration Filter (CNN-ItemCF), and the former has the fastest convergence speed, requiring only 40 epochs. The results show that the combination of FCNN and ItemCF algorithms can significantly improve the recommendation accuracy of college football course training strategies, ensure high training efficiency, and promote the football technical level of college students.