飞机机翼表面损伤的近红外高光谱识别

Near Infrared Hyperspectral Identification of Surface Damage on Aircraft Wings

飞机表面损伤的检测和设备研究,对于飞行安全与运营效益具有重大的现实意义。光谱匹配技术是飞机表面损伤高光谱检测需要解决的关键技术之一。不同光谱匹配算法的识别精度常因研究对象而异。为了利用光谱匹配算法实现飞机样品损伤识别,首先搭建了室内近红外高光谱飞机表面损伤检测系统,采集了参考样品和蒙皮样品两类样品的高光谱数据,并利用损伤像元光谱和无损像元光谱制作了两类像元的标准光谱。接着基于计算待测像元光谱与标准光谱相似度的匹配方法,先后采用光谱角(SA)、马氏距离(MD)、光谱信息散度(SID)、光谱相关系数(SCC)四类单一光谱匹配算法和六类组合光谱匹配算法进行两类样品的损伤识别。利用总体分类精度Pa和Kappa系数对上述光谱匹配算法的损伤识别结果进行精度评价。通过对SA、MD、SID、SCC单一算法的阈值参数优化,给出能够较好满足检测需求的合理阈值组。进一步,基于四类单一算法进行组合匹配算法设计,分别采用SA-MD、SA-SID、SA-SCC、MD-SID、MD-SCC和SID-SCC六类组合算法对两类飞机样品进行损伤识别。结果表明,相比单一匹配算法,组合算法的整体识别准确率相对较高。最后,分别给出适用于飞机样品损伤识别的最优单一匹配算法和组合算法,其中SCC算法和MD-SCC算法,对两类样品的损伤识别率分别可达95%以上和97.5%,可为外场飞机表面损伤的高光谱检测提供技术支撑。

The development of detection techniques and equipment for aircraft surface damage has significant practical significance for flight safety and operational efficiency.Spectral matching technology is a crucial technology that must be addressed in the hyperspectral detection of aircraft surface damage.The recognition accuracy of different spectral matching algorithms often varies depending on the research object.To use a spectral matching algorithm to achieve damage identification of aircraft samples,this article first built an indoor near-infrared hyperspectral system for aircraft surface damage detection and collected hyperspectral data of reference samples and skin samples.It produced standard spectra of two types of pixels using damaged pixel spectra and non-destructive pixel spectra.Subsequently,based on the matching method for calculating the similarity between the measured pixel spectrum and the standard spectrum,four types of single spectrum matching algorithms,namely spectral angle(SA),Mahalanobis distance(MD),spectral information divergence(SID),and spectral correlation coefficient(SCC),and six types of combined spectrum matching algorithms,were used for damage identification of two types of aircraft samples.The accuracy of multiple spectral matching algorithms'damage identification results was evaluated using the overall classification accuracy P_a and Kappa coefficient.A reasonable threshold group that can better meet the detection requirements is provided by optimizing the threshold parameters of single algorithms,such as SA,MD,SID,and SCC.Furthermore,based on the above four types of single matching algorithms,six types of combined matching algorithms were designed and used for sample damage identification,like SA-MD,SA-SID,SA-SCC,MD-SID,MD-SCC,and SID-SCC.The results show that the identification accuracy of those combined algorithms is relatively higher than that of any matching algorithms.Finally,this article presents the optimal single matching algorithm and combination matching algorithm for damage identification of aircraft samples,with the SCC algorithm and MD-SCC algorithm achieving damage identification rates of over 95%and 97.5%for both types of samples,respectively.This can provide technical support for hyperspectral detection of aircraft surface damage in the outfield.