UAV Flight State Recognition Using AC-GAN Based Method

Unmanned Aerial Vehicle (UAV) has become a widely used tool in both military and civilian applications. As the UAV capabilities and technology complexity rapidly increase, UAV reliability and security are particularly essential. In order to improve the operational security, fault diagnosis, flight quality assessment and fuel consumption assessment are implemented. However, under different flight states, such as takeoff, climbing, level flight, etc. the possible failures, the meaning of flight quality and fuel consumption levels are different, so flight state recognition is the basis of the above work. Aiming at the problem that the flight time of various flight states is different, which leads to the imbalance of recognition samples, this paper proposes an Auxiliary Classifier Generative Adversarial Network (AC-GAN) based method. The method supplements samples by generating samples that enjoy the same distribution with the real data, thereby improving the accuracy of the recognition. It also segments the time series by sliding window method, to solve the classification problem of unequal time series. The performance of the proposed method is illustrated using actual UAV sensor data. The results show that the proposed method can effectively improve the recognition accuracy.