Aircraft Skin Damage Detection Method Based on Improved Mask Scoring R-CNN

At present, most of the non-destructive inspections of aircraft skins are visual inspections, which are prone to missed inspections due to the negligence of inspectors. Therefore, rapid and accurate detection of aircraft skin damage based on images is of great significance for improving the safety of the civil aviation industry. This paper proposes a damage detection method based on improved Mask Scoring R-CNN. First, the attention module is added to the feature extraction network to enhance the expression of features, thereby improving the effect of damage detection and segmentation. Secondly, according to the feature difference of different shapes and sizes of damage, the multi-scale pyramid feature fusion module of multiple paths is introduced to fuse more detailed shallow information in the deep features. Finally, a more effective classifier head structure is proposed, which further improves the detection effect of aircraft skin damage. Experiments show that compared with the original algorithm, the algorithm in this paper has improved 12.1 and 7.9 in the and two performance indicators, respectively.