基于前视钻孔图像的特征提取分类及全景图合成研究

发布时间：2019-06-11 23:49

【摘要】：地质勘探是地质工程前的必备过程,随着相关技术的发展,利用前视钻孔摄像技术获取钻孔孔壁图像进行地质分析已成为地质勘探中一项非常重要的技术。当前对前视钻孔图像的分析处理主要存在两方面问题,一方面,大量的钻孔图像分析分类工作给技术人员带来了较大的挑战;另一方面,前视钻孔图像难以合成孔壁平面全景图,限制了分析水平。利用数字图像处理及模式识别相关技术代替人工完成分类以及合成孔壁全景图,对于提高地质分析水平,拓展前视钻孔摄像的应用范围具有重要意义。本文以前视钻孔图像为研究对象,研究了前视钻孔图像的特征提取分类与全景图合成技术。在前视钻孔图像的特征提取和分类技术方面,本文以传统的特征提取方法为切入点,介绍了常用的特征提取方法,并分析了其存在的不足,重点介绍了Contourlet变换以及非下采样Contourlet变换的基本原理及特点,以及其在自然纹理特征上的表述优势,通过提取非下采样Contourlet变换子带系数的统计特征结合钻孔图像的Hu不变矩特征,再利用方差统计进行特征选择,构成特征向量,最后利用BP神经网络进行分类实验验证,得到了比较满意的实验结果。在前视钻孔图像全景图合成技术方面,本文分析了前视钻孔成像与数字光学成像的基本原理及成像特点,确定了前视钻孔图像合成孔壁全景图的可行性,经过圆心定位、环形区域展开和图像匹配拼接等关键技术的研究,得到了良好的钻孔孔壁平面全景图,为后续的地质分析奠定了基础。本文提出的算法均在Matlab平台进行了验证,得到了比较满意的结果,证明了算法的可行性。
[Abstract]:Geological exploration is a necessary process before geological engineering. With the development of related technologies, it has become a very important technology in geological exploration to obtain drilling hole wall images by using forward looking drilling imaging technology for geological analysis. At present, there are two main problems in the analysis and processing of forward-looking drilling images. On the one hand, a large number of drilling image analysis and classification work has brought great challenges to technicians; on the other hand, forward-looking drilling images are difficult to synthesize hole wall panoramic images, which limits the analysis level. It is of great significance to use digital image processing and pattern recognition technology to complete classification and synthesize panoramic images of hole wall instead of manual classification, which is of great significance to improve the level of geological analysis and expand the application range of forward looking drilling camera. In this paper, the drilling image is regarded as the research object, and the feature extraction classification and panoramic image synthesis technology of forward looking drilling image are studied. In the aspect of feature extraction and classification technology of forward looking drilling image, this paper introduces the common feature extraction methods based on the traditional feature extraction method, and analyzes its shortcomings, with emphasis on the basic principle and characteristics of Contourlet transform and non-downsampling Contourlet transform, as well as its advantages in natural texture features. By extracting the statistical features of non-downsampled Contourlet transform subband coefficients combined with the Hu invariant moment features of drilling images, and then using variance statistics to select features to form feature vectors, finally, the classification experiments are verified by BP neural network, and satisfactory experimental results are obtained. In the aspect of panoramic image synthesis technology of forward-looking drilling image, the basic principle and imaging characteristics of forward-looking drilling imaging and digital optical imaging are analyzed in this paper, and the feasibility of synthesizing hole wall panoramic image from forward-looking drilling image is determined. through the research of key technologies such as center positioning, annular area expansion and image matching and stitching, a good panoramic image of hole wall is obtained, which lays a foundation for subsequent geological analysis. The algorithms proposed in this paper are verified on Matlab platform, and satisfactory results are obtained, which proves the feasibility of the algorithm.
【学位授予单位】：山东科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P624;TP391.41

【参考文献】