Flight action recognition is of great significance to evaluate the quality of flight training and improve the pilots driving skills.In the flight action sequence data,the flight state data at a certain time and some flight parameter data corresponding to the state make a very important contribution to flight action recognition,but the traditional attention mechanism only focuses on the contribution value of the upper-layer state features and ignores the influence of the lower-layer features.In order to effectively extract the key feature representation of flight parameter data,the Focus Attention (FA) mechanism is proposed,which extends the traditional attention mechanism,further focuses attention and learns the contribution of its lower-layer features to the upper layer.At the same time,the FA mechanism is extended to the BiLSTM network and the FA-BiLSTM network model is proposed.The model not only focuses on the flight state data at the critical time in the flight action sequence,but also can learn the key flight parameter data in the flight state.Experiments show that this method effectively improves the accuracy of flight action recognition,and the weighted average accuracy reaches 94%.