In order to improve the efficiency of football special ability training, this paper applies deep learning algorithm to football training action recognition, proposes an improved DCNN method based on vector weighted attention mechanism, and successfully applies it to the key frame image feature extraction and recognition of football action videos. The experimental results show that this method is effective in football technical action recognition. In particular, it significantly improves the recognition accuracy in terms of the recognition error of action attitude angle. This achievement not only reflects the effectiveness of the vector weighted attention mechanism in improving the feature extraction capabilities of the network model, but also further proves the great potential of deep learning methods in the field of complex motion recognition. Therefore, it helps coaches and athletes analyze technical movements more scientifically, optimize training methods, and improve training effects. At the same time, it also provides reference and reference for action recognition research in other sports.