飞机蒙皮损伤红外图像分割方法研究

发布时间：2019-01-05 18:18

【摘要】：蒙皮是飞机机身的重要组成部分。对受损蒙皮的检测和处理,将决定飞机的安全运行与否。红外无损检测方法能将表面及内部缺陷均检测出来而不破坏蒙皮,为了识别和分析检出的缺陷目标,提取目标特征并建立损伤数据库以便未来更高效准确地分析蒙皮损伤,首要任务便是图像分割,研究它对量化分析和工程实践均有举足轻重的价值。本文基于以上目的,首先建立红外热成像检测系统,采集损伤蒙皮试件的红外图像,并选择合理的图像增强和去噪方法进行预处理。其次以传统图像分割方法为依据,就红外图像的缺点研究出两种新的红外图像分割方法。马尔科夫随机场(Markov Random Field,MRF)和改进梯度向量流(Gradient Vector Flow,GVF)主动轮廓(Active Contour Model,Snake)模型相博弈的方法中,使用最大类间方差法(OTSU)作初始分割,并行、独立地运行MRF区域分割和改进的GVF Snake边界提取,将两者的结果进行博弈,在冲突条件下寻找最佳策略,当达到纳什平衡时输出最优分割结果。结合SIR(Susceptible Infected Removed)模型的归一化随机游走(Random Walk,RW)方法中,借助均值漂移算法提高效率;受SIR传染病模型的启发,将细菌类比为图像上的像素点,推导出SIR模型下的随机游走算法,再作归一化处理,达到SIR扩散过程的稳定状态时即得最终分割。最后研究了所提两种方法在飞机蒙皮损伤红外图像分割中的应用。将前文得到的红外图像进行MATLAB仿真,定性分析各个相关方法的效果,并定量评价算法性能;再纵向对比两种方法的优缺点,对未来用于飞机原位检测后的图像处理很有价值。
[Abstract]:Skin is an important part of the aircraft fuselage. The detection and treatment of the damaged skin will determine the safe operation of the aircraft. Infrared nondestructive testing method can detect both surface and interior defects without destroying the skin. In order to identify and analyze the detected defect targets, extract the target features and establish a damage database to analyze the skin damage more efficiently and accurately in the future. The most important task is image segmentation, which plays an important role in quantitative analysis and engineering practice. Based on the above purpose, this paper first establishes the infrared thermal imaging detection system, collects the infrared image of the damaged skin specimen, and selects the reasonable image enhancement and denoising method to preprocess. Secondly, based on the traditional image segmentation method, two new infrared image segmentation methods are proposed. Markov random field (Markov Random Field,MRF) and modified gradient vector flow (Gradient Vector Flow,GVF) active contour (Active Contour Model,Snake) are used in the game, in which the maximum inter-class variance (OTSU) is used as the initial segmentation and parallel. The MRF region segmentation and the improved GVF Snake boundary extraction are run independently and the results of the two are gambled to find the best strategy under the conflict conditions and output the optimal segmentation results when Nash equilibrium is reached. In the normalized random walk (Random Walk,RW) method based on SIR (Susceptible Infected Removed) model, the mean shift algorithm is used to improve the efficiency. Inspired by the SIR infectious disease model, the bacteria are compared as pixels on the image, and the random walk algorithm under the SIR model is deduced, and then normalized to achieve the stable state of the SIR diffusion process, the final segmentation is obtained. Finally, the application of the two methods in infrared image segmentation of aircraft skin damage is studied. The infrared images obtained above are simulated by MATLAB, the results of each related method are qualitatively analyzed, and the performance of the algorithm is evaluated quantitatively, and the advantages and disadvantages of the two methods are compared longitudinally, which is very valuable for future image processing after in-situ detection of aircraft.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：V267;TP391.41

【参考文献】