Damage Detection And Recognition Of Aircraft Surface Image Using Deep Network

With the serving life of a large number of aircraft gradually increased,the corrosion resistance of them will decline.The use of aircrafts shows that the main form of its malfunction is structural damage,which is caused by cracks,corrosion and other damages.The aircraft with serious structural malfunction must be grounded and conducted with rigorous maintenance.Therefore,to ensure the safety and normal use of the aircraft in service,the non-destructive test is significantly important,meanwhile the aircraft's non-destructive test mainly focuses on the surface inspection of its skin.At present,the surface inspection of aircraft skin mainly based on the visual inspection that conducted by ground crew.This method not only costs large working intensity,long testing period,but also ends with an obvious omission factor.How to detect the surface damage of aircraft skin with high efficiency and accuracy has become a topic and one of the studying focuses.Aiming at the problems above,deep neural network has been used to detect the damage of aircraft skin surface in this paper.Based on consulting relevant literature at home and abroad,this paper studies the damage detection of the surface image of aircraft skin and the application in identification.Specific work described as follows:Focus on studying the research status quo of the non-destructive testing of aircraft skin,as well as common damage types and causes of aircraft skin.According to the requirements of damage detection of aircraft skin,this paper defined eight types of injuries and the image acquisition specification was made based on the requirements of sample quality.A large number of aircraft surface images that meet this acquisition specifications are obtained from multiple sources.The original image is processed through image cropping,image annotation,data augmentation and so on,two sample datasets for experiment are constructed respectively,image patch dataset and image annotation dataset.For deep learning,this paper focuses on two of the most mainstream model,the convolution neural network image classification model and the object detection algorithm based on deep learning.In this paper,the network structure and training process of these two models are introduced in detail.These two kinds of deep learning models are respectively constructed in this paper and applied to the damage detection of aircraft surface.The convolution neural network learns the feature information in the image patch dataset to classify the four types of skin surface damage.The object detection model that trained on the image annotation dataset is used to locate the damaged region and identify the damage type.Experiments show that the deep learning algorithm is suitable for the damage detection and recognition of the aircraft skin surface images,it can obtain higher recognition rate and detection speed.This paper innovatively introduces deep learning technology into the damage detection of aircraft skin surface,which can improve the detection level of aircraft skin damage and shorten the daily maintenance time of aircraft,which is of great significance to the flight safety.