An Aircraft Landing Gear State Detection and Recognition Method based on YOLO-V4

To solve the problems that most landing gear state detection methods based on image recognition or detection technology need to be divided into in multiple stages, which leads to low efficiency and weak image generalization ability for complex background and different acquisition angles, an single-stage aircraft landing gear state detection and recognition method based on YOLO-V4 is proposed. By adding landing gear state labels to aircraft samples, adding landing gear state output channels to the network, and cleverly designing the loss functions for training. The proposed method makes the detection results of visible landing gears in the corresponding viewing angle and the landing gear state of the aircraft object can be obtained at the same time, which has stronger generalization ability for images from different viewing angles and is easier to be applied in engineering. The experiment on self-made aircraft dataset shows that the proposed method is not only slightly higher than the two-stage method in the average accuracy rate of aircraft landing gear state recognition, but also the average accuracy of aircraft and landing gear detection is 98.92% and 76.00% respectively, which meets the requirements of practical application.