图像分块下的隧道裂缝识别方法

针对隧道衬砌表面不均匀光照、渗水和噪声等强视觉干扰，设计了基于图像分块的隧道衬砌裂缝检测算法；根据中国西部地区的地理特征和隧道衬砌的外观病害，研制开发出一种快速、自动化的非接触式智能隧道结构物外观检测系统; 以非均匀光照下隧道图像数据集为研究对象，在图像分块的基础上提出一种适用于隧道裂缝特征提取的图像识别算法；研究了电子元件产生的噪声，并分析和总结了隧道衬砌的灾害特征；根据裂缝特征和分辨率将图像矩阵划分为适当数量的区域块，根据区域块的灰度特征将原始图像划分为目标背景区、目标病害区、病害背景区和其他区域，通过最大类间方差法和局部阈值法分割得到了隧道裂缝的粗图像，在此基础上进行了粗图像裂缝特征提取；对原始图像的每个区域块进行了对比度受限的自适应直方图均衡操作和局部阈值分割，得到了细节图像；将细节图像和粗图像的重叠区域设为理想裂缝二值化图像；结合隧道结构物外观检测系统对不同方向的裂缝图像进行了二值化试验，并通过隧道裂缝定位和投影法得到了隧道衬砌图像中裂缝的位置信息和方向。研究结果表明:提出的算法对隧道裂缝识别的准确值、召回率和F值可分别达90.34%、98.78%和94.37%，既可以保证隧道裂缝的完整性，也可以在非均匀光照下最大程度地保留目标裂缝的细节，可用于处理一般灰度图像的二值化问题。 

Tunnel crack recognition method under image block

Considering the strong visual interference such as the non-uniform lighting, water seepage, and noise in tunnel lining surfaces, a crack inspection method based on image blocks was designed for tunnel lining. A rapid and automatic non-contact intelligent inspection system for the tunnel structure appearance was developed according to the geographical characteristics in West China and the appearance diseases of tunnel lining. With the tunnel image data set under the non-uniform lighting as the research object, an image recognition algorithm for the characteristic extraction of tunnel cracks was proposed on the basis of image blocks. The noise generated by the electronic components was studied, and the hazard characteristics of tunnel lining were analyzed and summarized. The image matrix was divided into an appropriate number of area blocks in view of crack characteristics and resolution, and the original image was divided into the target-background area, target-disease area, disease-background area, and other areas according to the grayscale characteristics of these area blocks. Then, the rough image of tunnel cracks was obtained by the maximum inter-class variance method and the local threshold segmentation. On this basis, the crack characteristics of the rough image were extracted. Each area block of the original image was subjected to a contrast limited adaptive histogram equalization and a local threshold segmentation for a detailed image. The overlapping area of the detailed image and the rough image was set as the ideal image of crack binarization. On the basis of the inspection system for the tunnel structure appearance, binarization tests were carried out on the crack images in different directions, and the location information and directions of cracks in tunnel lining images were obtained by the positioning of tunnel cracks and the projection method. Research results reveal that the accuracy, recall, and F value of tunnel cracks under the proposed algorithm can reach 90.34%, 98.78%, and 94.37%, respectively. The proposed algorithm can not only ensure the integrity of tunnel cracks, but also retain the details of target cracks to the maximum extent under the non-uniform lighting. Thus, it can be used to deal with the binarization problem of general grayscale images. 1 tab, 13 figs, 37 refs.