Airport flight area target real-time detection method based on multi-scale feature decoupling

The invention discloses an airport flight area target real-time detection method based on multi-scale feature decoupling, and the method comprises the steps: 1, obtaining an airport flight area monitoring video, and constructing an airport flight area target detection data set; 2, constructing a multi-scale feature fusion module to realize detection of targets of different scales; 3, classifying and positioning tasks in target detection are decoupled, and a learning network based on feature decoupling is constructed; and 4, adding a learning network based on feature decoupling and a multi-scale feature fusion module into a YOLOv5 target detection network, and training and optimizing a target detection model in combination with a loss function. According to the method, shallow detail information and deep semantic information are aggregated by using the multi-scale feature fusion module, the detection capability of targets of different scales is enhanced, classification and positioning tasks are decoupled by us