Research on intelligent aircraft recognition in flight command scenarios by fusion of deep CNN and object detection

The operating environment of aircraft is highly dynamic, including sudden weather changes, electromagnetic interference, and multi-target overlap in dense airspace, which leads to blurry depth features of the aircraft and makes it difficult to accurately obtain the level of feature changes that reflect the deep level features of the aircraft relative to the initial state, thereby reducing the accuracy of recognition. To this end, research is being conducted on intelligent recognition methods for aircraft in flight command scenarios that integrate deep CNN and object detection. Preprocessed images are input into the network to extract multi-scale deep level features, which are processed and fused by different modules to set constraints for unmanned aerial vehicle no fly zones. Modeling with overall component feature blocks, representing feature change levels using feature weights, and setting multi-task loss functions to achieve fast, accurate, and stable identification of aircraft in complex scenarios. The results show that the design method can recognize aircraft of all categories well, and the data deviation in speed recognition is controlled within ± 0.5%, proving that it has certain advantages in recognition accuracy.