Infrared Image Recognition Of Damage In The Airplane Skin Based On Machine Learning

The aircraft skin contact with the air directly, it is very easy to be damaged under the outer force for a long time in the process of flight. At present, researchers have already applied a variety of detection methods for the skin damage, however, the research on how to identify the specific types of internal and external damage more rapid and more accurate is imperfect. In this thesis, the infrared thermal imaging technology is used to nondestructive test for the skin damage of civil aviation aircraft. In order to achieve the classification and recognition of the five typical damages, two kinds of classification algorithm based on machine learning have been established through the second characteristic of infrared thermography.Firstly, an analysis is made for the common damages of aircraft skin and causes of them,and then summarizes a number of nondestructive testing technologies for skin damages in the domestic and overseas now, which leads to the infrared thermal imaging testing technology. In this thesis, the temperature data of aircraft skin is collected by the infrared thermal imager,and the damage characteristics of infrared thermography are extracted from the temperature matrix. At the same time, the characteristics of different damages could be distinguished by the parameters of gray level co-occurrence matrix.Secondly, two kinds of classification and recognition algorithm are applied to the damage identification of aircraft skin. The fuzzy membership function is introduced to support vector machine to ensure the accuracy of multiple classification problems, and the classification method is one against one with voting, what's more, the classification results are reflected on the software interface. Another algorithm, the damage image is made a singular value decomposition by symplectic group classifier based on Lie group machine learning, and then reconstructs the damage image, thereafter the category of skin damage would be determined according to the projected length that the reconstruction image was projected onto the original one.At last, classification of the two kinds of algorithms for the skin damage are analyzed,the results show that both of them could identify the skin damage accurately and effectively,what's different, the accuracy of support vector machine is higher under the condition of less training samples, but with the increase of training samples, the symplectic group classifier has more advantages.