In track and field training, precise motion analysis is a crucial learning process that plays a vital role. Traditional training methods lag behind in capturing motion features, leading to inconsistencies in performance evaluation. To enhance the accuracy of motion correction while improving detection efficiency, this study proposes a deep learning-based system designed to help track and field athletes optimize their technical movements. The system optimizes the CNN network structure and combines batch normalization technology to accurately extract and identify erroneous movements during training, providing athletes with immediate feedback and targeted correction suggestions. The experimental findings indicate that all four TPR parameters outperform the other two methods, with the detection accuracy of this approach exceeding 97%, thereby confirming the exceptional effectiveness of deep convolutional neural networks. This system can achieve high recognition accuracy in different sports events, effectively reducing the problem of human error in traditional teaching methods. The system functions include real-time image acquisition, deep learning analysis, intelligent scoring, and online guidance, providing more efficient and intelligent technical support for physical education teaching and training.