The existing system for dynamic motion analysis of track and field athletes faces the limitation of a single data source, which makes it impossible to fully and accurately capture the dynamic performance and details of athletes, affecting the accurate evaluation of training and competition. Traditional methods rely too heavily on a single modality of data, making it difficult to capture a holistic view of an athlete's movements, which in turn leads to suboptimal training adjustments. This paper introduces a multimodal neural network (MMNN) and an attention mechanism to effectively integrate and jointly decode the dynamic motion analysis system of track and field athletes based on a joint decoding strategy. This can make up for the shortcomings of various data sources, reduce information loss, solve the problem of inconsistent information between different data sources, and improve the accuracy and robustness of the system, thereby providing decision support and personalized training plans for athletes. The experimental results show that the introduction of the attention mechanism has a lower loss value and is more effective than the comparative graph convolutional network strategy system. The error value fluctuates between 0.5 and 1, which can better reduce the MSE and achieve better accuracy, thereby formulating a good training plan for athletes, strengthening the training status of athletes, and improving their competitive level.