Football Action Recognition Based on Explainable Artificial Intelligence Assisted by Uncrewed Aerial Vehicles

This study explores UAV-assisted methods for football action recognition, and integrates sensing technology with explainable artificial intelligence (XAI) to improve accuracy and real-time performance. This approach supports football tactical analysis and action optimization. Firstly, this study collects real-time three-dimensional (3D) motion data of players using multimodal sensor devices mounted on UAVs. The data is then standardized and enhanced to ensure high-quality input for the model. Secondly, based on XAI technology, an Attention Mechanism Enhanced 3D Convolutional Long Short-Term Memory (AM-E3D-LSTM) algorithm-based football action recognition model is proposed. The E3D-LSTM module first extracts spatiotemporal features from the data. It captures local spatial information using 3D convolution and models long-term temporal dependencies with LSTM units. The introduced AM dynamically adjusts the model’s focus on different time steps or spatial regions. This reduces the interference of noise and redundant information, enabling the model to capture key features related to specific actions more accurately. Finally, the models performance is analyzed. The results reveal that compared with the traditional LSTM algorithm, the accuracy of the proposed model is improved by 7%. Moreover, the method has excellent performance in real-time action analysis, and the recognition delay is controlled within 100 milliseconds, which meets the real-time analysis requirements. Therefore, the model proposed here performs excellently in football action recognition, especially in processing complex action sequences. The proposed model shows high accuracy and stability and provides strong technical support for real-time football tactical analysis and sports performance optimization.