End-To-End Aircraft Skin Damage Detection Method Based on Ghostnet

The traditional aircraft skin inspection is mainly manual inspection, which has low efficiency, large workload, and is prone to missed and misdetected inspections. In order to improve the efficiency of aircraft skin detection, an end-to-end aircraft skin damage detection method based on Ghostnet is proposed. By introducing a scale factor to adjust the convolution method in Ghostnet, the multi-scale feature extraction module extracts the texture features of aircraft skin, which increases the range of receptive fields in the backbone network. Secondly, a multi-layer feature fusion module is introduced to integrate shallow and deep features. The angular margin is introduced to improve the confidence function to improve the confidence of each damage category. Based on the self-made data containing 1730 pieces of aircraft skin damage, the detection accuracy of the model can reach 89.28%, the detection speed is 36 frames per second, and the detection accuracy is improved by 9.28%.