基于小波神经网络的地面三维激光扫描点云数据的滑坡监测研究
发布时间：2019-04-12 18:24
【摘要】：地面三维激光扫描在获取面状数据领域发挥的作用越来越大,其地位越来越高。用它采集滑坡区域的点云数据时,必然要求能够精确地求得滑坡区域的沉降值;然而要想在硬件上取得突破,一般是比较难的。因此我们把精确地获得滑坡区域沉降值的这一目标放在对它的数据处理上来。鉴于许多学者已使用了时间序列或者卡尔曼滤波等方法求取滑坡区域的沉降值,本文采用小波神经网络分析方法实现精确地求得滑坡区域的沉降值。研究内容主要包括地面三维激光扫描数据的预处理部分和利用小波神经网络的分析方法进行网络训练并且最终精确求得沉降值两部分。地面三维激光扫描数据的预处理包括点云的拼接、滤波、缩减和分割四部分,其最终实现了把不同测站测得的点云坐标统一在同一个坐标系下,剔除了粗差比较大的点云。基于小波神经网络分析方法进行曲面拟合,以及精确地求取滑坡区域的沉降值,首先利用了隐式曲面构造的基本原理,采用了双三次插值拟合的方法构造出了关于点云数据的隐式曲面,发现采用这种方法拟合出的曲面精度不高,因此这种方法不适合针对大规模点云数据进行曲面拟合。采用小波神经网络分析方法,经过网络训练后的点云数据就可以对这个神经网络进行非线性的无限逼近,进而得到一个的隐式曲面;与此同时,也用BP神经网络对点云数据进行了曲面重建,通过对比两种网络模型的曲面拟合精度,发现小波神经网络具有比基于BP神经网络的地表模型拟合表现出更高的精度,更高的收敛速度;利用MATLAB软件平台编写了人机交互图形化用户使用界面GUI,求得了沉降值,验证了利用小波神经网络对曲面进行拟合这种方法的可靠性和科学性。
[Abstract]:3-D laser scanning plays a more and more important role in obtaining surface data, and its position is getting higher and higher. When it is used to collect point cloud data of landslide area, it is necessary to obtain the settlement value of landslide area accurately, however, it is difficult to make a breakthrough in hardware. Therefore, we put the goal of accurately obtaining the settlement value of landslide area on the processing of its data. In view of the fact that many scholars have used time series or Kalman filter to calculate the settlement value of landslide region, wavelet neural network analysis method is used in this paper to obtain the settlement value of landslide region accurately. The research includes two parts: the preprocessing part of the ground 3D laser scanning data and the training of the network by using the wavelet neural network analysis method and the final accurate calculation of the settlement value. The preprocessing of ground 3D laser scanning data consists of four parts: point cloud splicing, filtering, reduction and segmentation. Finally, the point cloud coordinates measured by different stations are unified in the same coordinate system, and the point clouds with large gross errors are eliminated. Based on wavelet neural network analysis method, surface fitting is carried out, and the settlement value of landslide area is calculated accurately. Firstly, the basic principle of implicit surface construction is used. The implicit surface of point cloud data is constructed by using bicubic interpolation fitting method. It is found that the accuracy of surface fitting by this method is not high, so this method is not suitable for surface fitting of large-scale point cloud data. Using wavelet neural network analysis method, the point cloud data after network training can be nonlinear infinite approximation to this neural network, and then an implicit surface can be obtained. At the same time, BP neural network is used to reconstruct the surface of point cloud data. By comparing the surface fitting accuracy of the two network models, it is found that the wavelet neural network has higher accuracy than the surface model fitting based on BP neural network. Higher convergence rate; The settlement value of graphical user interface (GUI,) for human-computer interaction is obtained by using MATLAB software platform, and the reliability and scientificity of this method for surface fitting using wavelet neural network is verified.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P225.2
本文编号：2457265
[Abstract]:3-D laser scanning plays a more and more important role in obtaining surface data, and its position is getting higher and higher. When it is used to collect point cloud data of landslide area, it is necessary to obtain the settlement value of landslide area accurately, however, it is difficult to make a breakthrough in hardware. Therefore, we put the goal of accurately obtaining the settlement value of landslide area on the processing of its data. In view of the fact that many scholars have used time series or Kalman filter to calculate the settlement value of landslide region, wavelet neural network analysis method is used in this paper to obtain the settlement value of landslide region accurately. The research includes two parts: the preprocessing part of the ground 3D laser scanning data and the training of the network by using the wavelet neural network analysis method and the final accurate calculation of the settlement value. The preprocessing of ground 3D laser scanning data consists of four parts: point cloud splicing, filtering, reduction and segmentation. Finally, the point cloud coordinates measured by different stations are unified in the same coordinate system, and the point clouds with large gross errors are eliminated. Based on wavelet neural network analysis method, surface fitting is carried out, and the settlement value of landslide area is calculated accurately. Firstly, the basic principle of implicit surface construction is used. The implicit surface of point cloud data is constructed by using bicubic interpolation fitting method. It is found that the accuracy of surface fitting by this method is not high, so this method is not suitable for surface fitting of large-scale point cloud data. Using wavelet neural network analysis method, the point cloud data after network training can be nonlinear infinite approximation to this neural network, and then an implicit surface can be obtained. At the same time, BP neural network is used to reconstruct the surface of point cloud data. By comparing the surface fitting accuracy of the two network models, it is found that the wavelet neural network has higher accuracy than the surface model fitting based on BP neural network. Higher convergence rate; The settlement value of graphical user interface (GUI,) for human-computer interaction is obtained by using MATLAB software platform, and the reliability and scientificity of this method for surface fitting using wavelet neural network is verified.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P225.2
【参考文献】
相关期刊论文 前2条
1 张邦礼，李银国，曹长修;小波神经网络的构造及其算法的鲁棒性分析[J];重庆大学学报(自然科学版);1995年06期
2 吕朝霞,胡维礼;小波网络在控制系统中的应用[J];信息与控制;2000年06期
相关硕士学位论文 前5条
1 周华伟;地面三维激光扫描点云数据处理与模型构建[D];昆明理工大学;2011年
2 沈剑;三维激光扫描重建技术探讨与分析[D];东华理工大学;2012年
3 朱云福;基于三维激光扫描数据的岩体结构面识别方法研究及系统研制[D];中国地质大学(北京);2012年
4 黄飒;三维激光扫描技术应用于古建筑测绘及其数据处理研究[D];河南理工大学;2012年
5 代修波;手持式相位激光测距仪的研究与设计[D];扬州大学;2013年
，本文编号：2457265
本文链接：https://www.wllwen.com/kejilunwen/dizhicehuilunwen/2457265.html
