有砟铁路路基病害的雷达图像识别方法研究

发布时间：2018-07-03 12:18

本文选题：探地雷达 + 有砟铁路　；参考：《北京交通大学》2016年博士论文

【摘要】：目前我国铁路总里程突破11.2万km,有砟铁路(包含桥梁和隧道)总里程近9.6万km。针对有砟铁路的里程数和状态检测要求,探地雷达技术成为当今连续检测有砟铁路路基的主要技术手段。然而,探地雷达检测数据量大,处理识别时间延迟,致使数据解释和病害识别存在主观性强、判别标准不一、识别精度低等诸多问题,因此,亟需开展有砟铁路路基病害的雷达图像识别方法研究,实现路基病害的快速、准确识别,为病害的整治处理提供技术支持。本文针对铁路路基病害识别过程中面临的判别标准不一、时效性差、识别精度低等诸多问题,依托国家"863"计划项目(2009AA11Z212)和河北省应用基础研究重点资助研究项目(11963544D),以大秦铁路的部分路段为研究对象,通过实际检测数据和理论分析的研究,建立了基于雷达图像特征的典型路基病害分类方法,分析了典型路基病害(缺陷)的发生机理、发育特征和空间分布特性,进而研究了典型路基病害的参数表征以及识别方法,最终形成了基于探地雷达技术的铁路路基病害快速识别方法。本文主要研究成果和结论如下：1、在充分认识现行路基病害分类方法的基础上,分析了典型路基病害的表征现象、基本成因和形状特征,并结合探地雷达技术的检测特点和铁路路基在雷达图像的表征形式,建立了基于雷达信号特征的路基病害分类方法,为路基病害的特征提取以及识别技术提供前期基础。2、基于多条运营线路路基的雷达实测数据,分析了典型路基病害的发生机理及空间分布,得出了相应病害在长度及深度方向上的统计分布规律,为后续的病害特征提取及随机匹配提供了必要的先验信息支持。根据铁路路基检测技术指标以及探地雷达的技术参数,设计了满足路基检测要求的探地雷达工作参数及雷达天线配置,并将系统安装于轨道检查车上,采用频率为100M和400M雷达天线,同时检测不同深度的路基状态,完成了大秦线部分路段路基的检测。3、本文从时域、时频域和物理几何稀疏性等方面对典型路基病害雷达图像的特征提取进行研究,并建立了最优的雷达信号特征表示。一种是通过主成分分析确定了以分块能量、分层点频率和能量均值作为铁路路基的时域特征值,获得了较低维的空间特征：一种是对典型路基病害雷达图像的距离-深度二维时频特征表现进行了分析,获取了路基雷达检测图像的低频信息和小波多尺度空间细节能量谱特征,从而减少了采样数据量；一种采用稀疏表示的方法提取雷达信号特征向量,采用投影变换得到了路基病害的垂直和水平投影线段,并以此作为判断路基病害类型的依据,从而为图像的特征提取提供了理论依据。4、针对反映路基雷达图像同相轴延续性和紊乱程度的时域特征存在的冗余性和局限性问题,通过主成分分析确定了低维时域特征值。同时在所确定时域特征的基础上,分别采用了人工神经网络、广义神经网络聚类算法和支持向量机等人工智能的方法对路基病害进行识别并加以比较,最终确定了以支持向量机作为路基病害的有效识别算法,从而为铁路路基及其典型病害的识别奠定了算法基础。5、利用路基病害的物理几何特点,通过提取有效的雷达图像的边界曲线,得到了路基病害的垂直和水平投影线段,并将垂直投影线段和能量谱作为路基病害的有效特征,最终形成了基于该特征的铁路路基病害的识别方法。分析结果表明：数据的采集量仅为原数据量的1/4,识别速度提高将近4倍,并提出了HS-SVM方法识别路基病害,且其识别率在85%以上,实现了路基病害的快速识别。
[Abstract]:At present, the total mileage of railway in China breaks through 112 thousand km, and the total mileage of ballasted Railway (including bridges and tunnels) is nearly 96 thousand km. for the mileage and state detection requirements of ballasted railway. Ground penetrating radar technology has become the main technical means for continuous detection of ballastballed railway subgrade. However, the detection data of the ground penetrating radar is large and the time of processing recognition is delayed. There are many problems in data interpretation and disease identification, such as strong subjectivity, different criteria and low recognition accuracy. Therefore, it is urgent to carry out the research of radar image recognition method for the disease of ballastballed railway subgrade, to realize the rapid and accurate recognition of subgrade diseases, and to provide technical support for the treatment of disease. According to the different criteria of discrimination, poor timeliness and low recognition accuracy, we rely on the national "863" project (2009AA11Z212) and the key support research project (11963544D) of the applied basic research in Hebei Province, and take the section of the railway of the Daqin Railway as the research object, and establish the basis of the research on the actual testing data and theoretical analysis. The typical subgrade disease classification method of radar image features analyses the occurrence mechanism, development characteristics and spatial distribution characteristics of typical subgrade disease (defect), and then studies the parameter characterization and recognition method of typical subgrade diseases, and finally forms a rapid identification method of railway subgrade disease based on ground penetrating radar technology. The results and conclusions are as follows: 1, on the basis of fully understanding the existing classification methods of subgrade disease, the characteristics of typical subgrade diseases, basic causes and shape characteristics are analyzed, and the detection characteristics of ground penetrating radar technology and the characteristic form of the railway subgrade in radar image are combined, and the subgrade disease based on radar signal features is established. This method provides the early foundation.2 for the feature extraction and recognition technology of subgrade diseases. Based on the radar data from the subgrade of multiple operating lines, the occurrence mechanism and spatial distribution of typical subgrade diseases are analyzed, and the statistical distribution rules of the corresponding diseases in the direction of length and depth are obtained, and the characteristics of the disease are extracted and random. The matching provides the necessary prior information support. According to the technical parameters of the railway subgrade detection and the technical parameters of the ground penetrating radar, the working parameters of ground penetrating radar and the configuration of the radar antenna are designed to meet the requirements of the subgrade detection, and the system is installed on the track inspection car, the frequency is 100M and 400M radar antenna, and the different depth is detected at the same time. The subgrade condition of the subgrade of the Daqin line is tested.3. This paper studies the feature extraction of the typical subgrade disease radar image from the time domain, time and frequency domain and the physical geometric sparsity, and establishes the optimal radar signal feature representation. One is to determine the block energy and the stratified point frequency by the principal component analysis. As the time domain characteristic value of the railway subgrade, the mean value of the energy is obtained, and the spatial characteristics of the lower dimension are obtained. One is the analysis of the feature of the distance to depth two-dimensional time-frequency characteristic of the typical subgrade disease radar image, and the low frequency information of the subgrade radar detection image and the feature of the wavelet multi scale spatial detail energy spectrum are obtained, thus reducing the mining of the subgrade. A method of sparse representation is used to extract the characteristic vector of radar signal, and the vertical and horizontal projection lines of subgrade disease are obtained by projection transformation, which is used as the basis for judging the type of subgrade disease, thus providing a theoretical basis for the feature extraction of the image, which reflects the continuity of the same phase axis of the subgrade radar image. At the same time, artificial neural network, generalized neural network clustering algorithm and support vector machine such as artificial neural network, generalized neural network clustering algorithm and support vector machine are used to identify the subgrade disease on the basis of the time domain characteristics. By comparing it, the effective recognition algorithm of support vector machine as subgrade disease is determined and the algorithm foundation.5 is laid for the identification of railway subgrade and its typical diseases. By using the physical geometric characteristics of the subgrade disease, the vertical and horizontal projection lines of the subgrade disease are obtained by extracting the effective boundary curve of the radar image. With the vertical projection line and the energy spectrum as the effective characteristics of the subgrade disease, the identification method of railway subgrade disease based on this feature is formed. The analysis results show that the data collection is only 1/4 of the original data, the recognition speed is nearly 4 times higher, and the HS-SVM method is proposed to identify the subgrade disease, and its recognition rate is 85% Above, the rapid identification of subgrade disease has been realized.
【学位授予单位】：北京交通大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】