基于图像处理的航空构件缺陷检测技术研究

发布时间：2018-02-02 06:45

本文关键词： 缺陷检测双向扩散模糊聚类算法二叉树分类器　出处：《南昌航空大学》2017年硕士论文　论文类型：学位论文

【摘要】：航空构件缺陷检测的主要目的是保证航空构件产品的质量与性能,避免因构件出现缺陷最终导致航空设备失效,出现安全事故。近年来,利用数字图像处理对各类缺陷进行自动检测的研究已经受到各国学者的重视。但是,由于缺陷图像边缘模糊、对比度较低等因素造成缺陷漏检、误判的现象,降低了缺陷的检测效率。因此,本文主要研究缺陷检测技术中的图像增强、图像分割与缺陷提取、缺陷特征描述与分类识别等内容。主要工作如下:(1)针对各向异性扩散模型对图像去噪增强时易出现边缘模糊、振铃效应等问题,提出了一种新的基于局部边缘特征的自适应双向扩散模型。该模型在四方向的梯度和拉普拉斯算子的基础上,首先对图像边缘的结构特征进行深入研究,以此区分出图像的平坦区域与边缘区域,然后针对各个区域的特征选择不同的扩散策略,使得该模型既能滤除噪声又能锐化图像的边缘细节。实验结果表明,与PM、CBDF和ABDF三种扩散模型相比,该模型能够有效地提升图像的对比度,增强缺陷的边缘细节。(2)由于传统模糊C均值聚类算法的模糊隶属度矩阵结构单一,且忽略了相邻像素的相关性,使得图像的分割结果不理想。虽然随后改进的各种聚类算法改善了分割效果,但运行效率偏低。因此,本文提出了一种基于分层的直觉模糊C均值聚类算法。首先,利用图像的隐含信息引入分层技术将图像划分为多个不同区域,计算其相应的聚类中心;然后,利用图像邻域像素之间的关联性,构造融合像素相关性和直觉模糊隶属度矩阵的目标函数,求解直觉模糊隶属度矩阵;最后,修正图像分层,直到符合聚类要求。实验结果表明,与FCM算法、FCMS算法和FLICM算法相比,本文算法不仅能够得到更好的图像分割效果,而且也较大地提升了图像分割效率。(3)针对航空构件常见缺陷的结构特征与分布情况,选取能够准确识别航空构件缺陷的特征参数,并采用二叉树分类器对缺陷进行分类识别。实验结果表明,本文的二叉树分类器能够基本识别航空构件的各类缺陷。
[Abstract]:The main purpose of aeronautical component defect detection is to ensure the quality and performance of aeronautical component products, to avoid the failure of aviation equipment and the occurrence of safety accidents due to component defects. The research on automatic detection of various kinds of defects by digital image processing has been paid more attention by scholars all over the world. However, due to the blur of the edge of the defect image and the low contrast, the defects are missed and misjudged. Therefore, this paper mainly studies image enhancement, image segmentation and defect extraction in defect detection technology. The main work is as follows: 1) aiming at the problems of edge blur and ring effect in anisotropic diffusion model for image denoising enhancement. A new adaptive bidirectional diffusion model based on local edge features is proposed. Based on the gradient of four directions and Laplace operator, the structural features of image edges are studied in depth. In this way, the flat region and the edge region of the image are distinguished, and then different diffusion strategies are selected according to the characteristics of each region, so that the model can not only filter the noise but also sharpen the edge details of the image. The experimental results show that the proposed model can not only filter the noise but also sharpen the edge details. Compared with the three diffusion models of PMDF-CBDFs and ABDF, this model can effectively improve the contrast of images. Because the traditional fuzzy C-means clustering algorithm has a single fuzzy membership matrix structure and neglects the correlation of adjacent pixels. The result of image segmentation is not ideal. Although the improved clustering algorithm improves the segmentation effect, the running efficiency is on the low side. In this paper, an intuitionistic fuzzy C-means clustering algorithm based on stratification is proposed. Firstly, the image is divided into several different regions by using the implicit information of the image, and the corresponding clustering centers are calculated. Then, the objective function of fusion pixel correlation and intuitionistic fuzzy membership matrix is constructed by using the correlation between adjacent pixels of image, and the intuitionistic fuzzy membership matrix is solved. The experimental results show that compared with FCM algorithm and FLICM algorithm, this algorithm can not only get better image segmentation effect. It also improves the efficiency of image segmentation greatly.) aiming at the structural features and distribution of common defects of aviation components, the feature parameters which can accurately identify the defects of aviation components are selected. The experimental results show that the binomial tree classifier can recognize the defects of aeronautical components.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：V267;TP391.41

【参考文献】