Civil aircraft cockpit abnormal behavior detection method based on cross-modal fusion

Civil aircraft monitoring scenes are complex and changeable, abnormal types are diversified, the difficulty of judging abnormity by using single modes such as videos, audios and Internet of Things information is high, the accuracy rate is low, the stability is poor, meanwhile, the occurrence probability of abnormal events is small, the types are diversified, sufficient sample data cannot be collected, and it is difficult to guarantee that under the non-controlled monitoring condition, the abnormal events cannot be detected easily. And the abnormal video behavior in the camera in the cockpit can be accurately judged. Aiming at the problems that the abnormal behaviors of the cockpit are diversified and single-modal detection cannot meet the existing requirements, the cross-modal fusion cockpit abnormal behavior detection method is adopted, and the invention provides the cross-modal fusion civil aircraft cockpit abnormal behavior detection method and device on the basis of fusing audio and video signals. A multi