同步碎石封层石子覆盖率检测系统算法研究

发布时间：2018-05-26 14:25

本文选题：同步碎石封层 + 颗粒图像几何畸变　；参考：《郑州大学》2015年硕士论文

【摘要】：同步碎石封层施工是一种先进的公路施工工艺,石子撒布覆盖率是同步碎石封层施工的一项重要指标。目前的石子撒布率检测方法依靠人眼观察或计数,工作量大、效率低、受到主观因素影响,并且难以融入自动化施工管理系统。而基于图像的工控和检测技术正在迅速发展,为了能够应用数字图像处理技术解决石子撒布率检测的问题,本文对有关算法进行了研究。本文首先研究了为实现石子撒布率检测系统所必须的图像预处理算法,主要包括图像几何畸变恢复和图像增强,接着研究并实验了多种形态学图像处理技术在施工石子颗粒图像的应用,最后,本文提出一种采用分水岭算法和区域形态学特征,对颗粒图像进行分割与合并,并对颗粒进行计数的算法。本文的主要工作可概括为以下几个方面:1、为了恢复镜头的几何畸变的图像,往往根据几何模型重新计算像素位置并采用双线性插值,由于一幅图像中不同位置的畸变程度不同(中间部分畸变较小,边缘部分畸变强烈),插值造成的图像模糊程度也不相同。本文根据其频域响应的分析,在图像的不同位置采用不同的图像增强参数,改善了上述问题。2、本文第二部分重点研究了形态学图像处理的算法,包括区域生长算法,基于标记的分水岭算法,以及基于距离变换(distance transform)的分水岭算法。通过实验可以比较出这些算法处理石子颗粒图像的效果。3、针对粘连严重、分辨率和清晰度较低的颗粒图像分割和计数的问题,本文提出了一种基于分水岭和形态学特征的方法。首先使用分水岭算法分割图像,得到过分割的结果,接着通过本文定义的区域形态学特征,采用加权马氏距离和区域连接图(region adjacency graph)指导过分割区域的合并。还讨论了加权马氏距离的性质,以及一些在不影响效果的情况下减少运算量的机制。通过实验表明,该算法效果良好,对原始图像的灰度、对比度和噪声变化具有不变性,在查全率-准确率(precision-recall)曲线的表现上优于现有方法。
[Abstract]:Synchronous gravel sealing construction is an advanced highway construction technology, and the coverage rate of stone distribution is an important index of synchronous gravel sealing construction. The present method for detecting the rate of stone distribution depends on the observation or counting of human eyes, which has the advantages of heavy workload, low efficiency, subjective factors, and difficult to be integrated into the automatic construction management system. But the technology of industrial control and detection based on image is developing rapidly. In order to solve the problem of detecting the distribution rate of stone by using digital image processing technology, the relevant algorithms are studied in this paper. In this paper, we first study the image preprocessing algorithm which is necessary to realize the detection system of stone distribution rate, which includes image geometric distortion recovery and image enhancement. Then we study and experiment the application of various morphological image processing techniques in the construction of stone particle image. Finally, this paper proposes a watershed algorithm and regional morphological features to segment and merge the particle image. And the algorithm of counting the particles. The main work of this paper can be summarized as follows: 1. In order to restore the geometric distortion image of the lens, the pixel position is often recalculated according to the geometric model and bilinear interpolation is used. Because the distortion degree of different positions in an image is different (the middle part distortion is small, the edge part distortion is strong), the image ambiguity degree caused by interpolation is also different. In this paper, according to the analysis of frequency domain response, different image enhancement parameters are adopted in different positions of the image to improve the above problem. The second part of this paper focuses on the morphological image processing algorithm, including the region growth algorithm. The watershed algorithm based on label and the watershed algorithm based on distance transform. The results of these algorithms can be compared by experiments. Aiming at the problem of segmentation and counting of grained images with serious adhesion and low resolution and sharpness, a method based on watershed and morphological features is proposed in this paper. Firstly, the watershed algorithm is used to segment the image, and the result of over-segmentation is obtained. Then, the weighted Markov distance and the region adjacency are used to guide the merging of the over-segmented regions through the morphological features of the region defined in this paper. The properties of weighted Markov distance and some mechanisms for reducing computation without affecting the effect are also discussed. The experimental results show that the algorithm is effective and invariant to the gray, contrast and noise changes of the original image, and is superior to the existing methods in the performance of recall-accuracy precision-recalline curve.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U416.2;TP391.41

【共引文献】