【作者】 冷彪；

【导师】 仇文革；

【作者基本信息】 西南交通大学 ， 桥梁与隧道工程， 2009， 博士

【摘要】 论文从研究隧道掌子面地质图像入手,通过数字图像处理技术对隧道掌子面地质图像进行分析和处理,以自动或半自动方式提取出掌子面上岩体层理、节理、断层等结构面信息。对宏观上较为明显的结构面,建立相邻隧道掌子面上各岩层层理、节理、断层等结构面的对应关系,从而根据三维建模技术建立隧道开挖部分的三维地质结构模型,并以此预测掌子面前方未开挖部分的岩体结构等信息。根据对前方未开挖部分的岩体结构的预测,反馈给隧道设计与施工,为实现隧道信息化设计与施工提供了重要的技术支撑。主要结论和成果如下：1、通过数字图像处理技术,对隧道掌子面地质图像进行图像预处理、边缘检测和边界提取,对同一结构面上不连续结构面边界采用边界聚合、边界自动连接算法进行连接,提取出的边界可作为地质素描图中的结构面边界,加以少量的人工干预,形成地质素描图,提高了地质素描速度,同时使人工地质素描中因素描人员不同结果也可能极不相同的情况得到有效控制。2、对于自动边界提取过程中出现实际结构面边界线漏检和误检的情况,加入人工智能剪刀功能。它将图像边界自动提取和人工干预的方式相结合,实现了对结构面边界的半自动提取,既解决了全自动边界提取过程中边界提取错误的情况,也减少了完全手工对结构面边界提取费时、费力且不能精确定位结构面边界的情况。3、通过数字图像处理技术,从细观上对隧道掌子面岩体特征进行了统计分析,取得岩体结构面长度、单位面积裂隙长度、结构面平均间距、RQD等结构面的自然特征,并据此对掌子面岩体状况进行描述。实现了对隧道掌子面岩体的自动分析和评价功能。4、对隧道掌子面附近边墙及拱部未衬砌区域的围岩,分析了图像采集时需要注意的问题,研究了生成地质展开图像的算法。5、通过数字图像处理技术,对近距离拍摄的多种岩石标本的大量岩石图像样本进行特征提取。根据BP神经网络算法,对图像特征参数进行训练,得到参数模型,该模型可用于自动判别不同的隧道掌子面岩体的岩性。6、根据现场测得的结构面产状信息,对自动建立相邻隧道掌子面上岩体层理、节理、断层等结构面边界线的对应关系进行了探讨。根据OpenGL三维建模技术及三维重建相关算法,建立隧道已开挖部分的三维地质结构模型,实现了隧道地质结构的可视化,从而更加直观的认识隧道开挖部分的地质情况。7、根据已开挖部分的三维地质结构模型,预测隧道掌子面前方未开挖部分的三维地质结构。由隧道三维地质结构模型及预测模型,生成隧道边墙及拱部的地质展开图、横剖切面、纵剖切面、虚拟地质钻孔等图形,同时利用结构面在大地坐标系下的空间位置关系,分析和计算出各结构面的岩层产状以及结构面走向与隧道开挖方向间的夹角。这些信息有利于全面分析和掌握隧道的地质状况。8、根据对隧道掌子面的分析,实现了对“隧道掌子面地质信息系统(TFGIS) 2008"的研发,为隧道信息化设计与施工提供了有力的技术支撑。更多还原

【Abstract】 Starting from studying tunnel face geological images, this paper analyzes and processes tunnel face geological images through digital image processing technology. Structural planes, such as rock mass bedding, joints and faults, are automatically or semi-automatically extracted. For those obvious structural planes, they are corresponded between neighboring faces so that the three-dimensional geological structure model of the tunnel having been excavated part is rebuilt according to three-dimensional modeling technology. From this model, the information of rock mass structures in front of current face is predicted. All this information is given to tunnel design and construction, so as to offer important technical support for realizing tunnel information design and construction.The main conclusions and achievements are as follows:1. By using digital imaging technology, face images are pre-processed, edge-detected and boundary-extracted. For a structural plane, if any of its’ boundaries are not continuous, boundary polymerization, auto boundary linking algorithm are introduced to connect this discontinuous boundary lines. These boundaries can be regarded as those in geological sketch. Once much manual intervention is applied, the boundary graphs can form geological sketchs. This will improve geological sketch speed. Correspondingly, the situation that manual sketch results are very different because of different sketch operators is efficiently controlled.2. Considering that undetected and false boundaries are often exist when structural plane boundaries are extracted automatically, manual intelligent scissor which combines auto extracting image boundary with manual intervention is introduced to semi-extracts structural plane boundaries. This method dismisses boundary extraction errors. And it also reduces boundary extraction time, energy, and inaccurate positioning. These cases often exist if completely manual intervention is adopted.3. Rock masse characters in faces are statistically analyzed by using digital image processing technology. The natural characters of structural planes, such as structural plane length, joint length of unit area, average distance between structural planes, RQD, are acquired. According to these data, the conditions of surrounding rocks in tunnel faces are described. So the function that automatically analyzed and evaluated face surrounding rock conditions is realized.4. For the surrounding rocks that are adjacent to faces and not laid lining, these questions that need to be cared during image collections are analyzed. And how to produce geological launched maps is researched.5. By digital image processing technology, characteristic extraction is applied to lots of rock image samples, which are gained through closely shooting many kinds of rocks. According to BP neural network algorithm, image characters are trained so as to obtain parameter model. This model can be used to automatically classify the lithology of different tunnel faces.6. In accordance with actually measured attitude information of structural planes, how to correspond with structural planes between neighboring faces is discussed. In the light of OpenGL three-dimensional modeling technology and correlatively three-dimensional rebuilding algorithms, the tunnel three-dimensional geological structural model is rebuilt, which means that the visualization of tunnel geological structure is realized. So the geological condition of having being excavated part of tunnel is acquainted more directly.7. Based on the three-dimensional geological structural model, the unexcavated part in front of current face is predicted. From the predicted model, geological launched map, cross-cutting surface, longitudinal cutting face and virtual geological drilling are built. At the same time, with the special coordinate relation of structural planes in geodetic coordinate system, all structural planes’attitudes and the angle between strike and the direction of tunnel excavation are analyzed and computed. All the information is useful to extensively learn about and analyze tunnel geological conditions.8. According to the analysis of tunnel faces, "Tunnel Face Geology Information System(TFGIS) 2008" has been developed, which provides strong technical support for tunnel information design and construction.更多还原