榆神矿区是我国主要煤炭生产基地之一，高强度地下采煤引起的地表沉陷对煤矿安全生产和矿区生态环境造成了严重影响。为了快速、高效地获取地表沉陷变形信息，除了采用常规观测站方式开展变形监测外，近年来InSAR、激光扫描、摄影测量等技术也应用于矿区地表沉陷监测，但都存在一定的局限性。其中，低空无人机航摄技术在构建精细地形模型时虽简便可行，但在模型构建和叠加过程中受植被等噪声影响，往往导致沉陷模型误差过大，制约了该技术在矿区沉陷监测中的实际应用。本文针对榆神矿区地表植被覆盖度低但沉陷量较大的特点，以小保当煤矿综采工作面开采沉陷区为研究对象，利用低空无人机多期航拍影像数据构建地表沉陷模型，实现全盆地沉陷信息的高效、精细提取。主要研究内容及结果如下：

（1）结合榆神矿区综采工作面开采地表沉陷特点设计了无人机航拍影像数据采集和处理的技术方案。基于影像空间分辨率要求确定了相应的外业航飞参数和像控点布设方案。在优化数据采集和处理技术流程的基础上，利用检查点对比分析了所获取影像数据在平面位置及高程上的实际精度，结果表明利用低空无人机航拍影像建模能够满足榆神矿区地形平缓条件下大采高开采沉陷盆地精细建模的基本要求。

（2）针对研究区条件下获取的影像采用多种滤波和插值算法进行点云数据处理，通过实验对比确定了适用于榆神矿区地貌和植被覆盖条件的点云滤波与插值算法。分别采用三角网渐进加密滤波、数学形态学滤波、移动窗口滤波、地形坡度滤波等算法去除非地面点云，发现三角网渐进加密滤波算法的效果最优，并确定了相应的最优迭代参数。通过对比分析四块实验区不同算法的插值精度，表明局部多项式插值算法所构建的DEM精度优于其它几种算法。

（3）利用机器学习分类算法对滤波后的植被点云进行后处理，进一步提升了点云去噪效果。分别采用人工神经网络法（ANN）、最大似然法(MLC)、支持向量机法(SVM)、随机森林法(RFC)进行分类实验，结果表明SVM法分类精度优于其他三种方法；在二次去除植被点后保留的地面点云占比明显提高，点云误差均值和中误差明显减小，有效地剔除了非地面信息的影响，提升了沉陷区DEM精细建模的实际精度。

（4）根据研究区多期无人机影像生成的DEM叠加构建了地表沉陷盆地的精细三维模型，通过实测数据验证了沉陷模型的实际精度，并实现了沉陷盆地的数学建模。将生成的沉陷模型分别与机载激光扫描监测结果和常规观测站获取的下沉曲线进行对比，验证了所构建的沉陷模型的精度。进一步利用概率积分法开采沉陷预计模型拟合下沉盆地，并确定了相应的模型参数，实现了航测沉陷模型的数学表达。

本文研究结果通过改进基于无人机影像数据构建矿区沉陷模型的技术流程提高了模型实际精度，有助于推进该技术方法在榆神矿区大范围、高强度采煤地表塌陷监测中的实际应用。

Yushin mining area is one of the main coal production bases in China. The surface subsidence caused by high intensity underground mining has seriously affected the safety production of coal mines and the ecological environment of the mining area. In order to obtain the surface subsidence deformation information quickly and efficiently, in addition to the conventional observation stations to carry out deformation monitoring, InSAR, laser scanning, photogrammetry and other technologies to carry out mining subsidence monitoring have also achieved some achievements in recent years, but there are certain limitations. Among them, the low-altitude uav aerial technology has been widely used in 3D modeling of real by superimposing multiphase image build fine terrain model for surface subsidence information, on technology is simple and feasible, but when the terrain modeling error caused by vegetation such as noise impact is too large, often restricted the practical application of this technique in coal mining subsidence monitoring. Aiming at the characteristics of yushen mining area with low surface vegetation coverage but large amount of subsidence, this paper takes the mining subsidence area of fully mechanized mining face of Xiaobaodang Coal Mine as the research object, and uses low-altitude UAV multi-period aerial image data to construct a subsidence model to realize efficient extraction of subsidence information of the whole basin. The main research contents and results are as follows:

(1) Combined with the characteristics of surface subsidence of fully mechanized mining face in Yushen Mining area, the technical scheme of data acquisition and processing of aerial photography by UAV was designed. Based on the requirement of image spatial resolution, the corresponding field flight parameters and image control point layout scheme are determined. Based on the optimization of data acquisition and processing technology process, the actual accuracy of the obtained image data in plane position and elevation is compared and analyzed by using checkpoints, which indicates that the modeling of low-altitude UAV aerial image can meet the basic requirements of fine modeling of the mining subsidence basin with large mining height under the condition of gentle terrain in Yushen Mining area.

(2) A variety of filtering and interpolation algorithms were used for point cloud data processing of the images acquired in the experimental area, and the point cloud filtering and interpolation algorithm suitable for the geomorphic environment of Yushen mining area was determined through experimental comparison. The algorithm of triangulation network progressive encryption filtering, mathematical morphology filtering, moving window filtering and terrain slope filtering were used to remove ground point cloud respectively. It was found that the effect of triangulation network progressive encryption filtering algorithm was the best, and the corresponding optimal iteration parameters were determined. By comparing and analyzing the interpolation precision of different algorithms in four experimental areas, it is shown that the DEM precision constructed by local polynomial interpolation algorithm is superior to other algorithms.

(3) Use machine learning classification algorithm to classify the images in the study area and post process the vegetation point cloud to further improve the point cloud denoising effect. Artificial neural network method (ANN), maximum likelihood method (MLC), support vector machine (SVM) and random forest method (RFC) were used for classification experiments. The results show that SVM method has better classification accuracy than the other three methods. After the secondary removal of vegetation points, the proportion of retained ground point cloud is significantly increased, and the mean and median errors of point cloud are significantly reduced, which effectively removes the influence of non-ground information and improves the actual accuracy of DEM fine modeling in subsidence area.

(4) Based on the DEM superimposed by multi-period UAV images in the experimental area, a fine 3D model of the surface subsidence basin is constructed. The actual accuracy of the subsidence model is verified by the measured data, and the mathematical modeling of the subsidence basin is realized. The subsidence model is compared with the 3d laser scanning monitoring results and the conventional observation line subsidence curve respectively to verify the accuracy of the constructed subsidence model. Furthermore, the probability integration method is used to fit the subsidence prediction model of mining basin, and the corresponding model parameters are determined, and the mathematical expression of aerial survey subsidence model is realized.

The research results of this paper improve the actual accuracy of the model by improving the technical process of constructing the mining subsidence model based on UAV image data, which is conducive to promoting the application of this technology method in the monitoring of large-scale and high-intensity mining surface subsidence in Yushen Mining area.