基于稀疏混合估计的坡度超分辨率重构方法研究

发布时间：2018-07-29 19:46

【摘要】：坡度(Slope)是地形特征的重要参数,也是地貌变化的重要评价指标。区域尺度上的坡度,通常由DEM(Digital Elevation Model)计算获得。由于高分辨率DEM数据获取较困难,而低分辨率下,坡度数据的高频和低频部分损失严重,因此通过超分辨率重构(降尺度变化)的方式来获取高精度坡度数据成为解决问题的途径之一。本研究通过改进的POCS(Projections Onto Convex Sets)算法和基于稀疏混合估计的方法对坡度数据的超分辨率重构进行研究,主要内容如下:(1)通过改进的POCS算法对坡度数据进行超分辨率重构。本算法首先针对读入的多组DEM数据进行判断,根据每个数据的方差结果来确定本组数据值的大致范围,对其中与平均范围差异较大的数据进行剔除,来获得更好的重构效果;利用双线性插值的方法来构造参考数据,使结果更加平滑真实;之后进行参考数据修正,修正过程是根据梯田坡度变化较大的值,将坡度数据进行边缘检测,获得检测结果后与将其对参考数据进行修正。(2)基于稀疏混合估计法对坡度数据进行超分辨率重构。算法首先将坡度数据、在近似、水平、垂直与45°四种方向上进行小波变换,该步骤可以充分提取坡度数据特征。然后建立坡度数据块字典和对块进行正交匹配追踪,通过构建坡度数据的稀疏冗余块字典,对坡度数据特性的归纳,使数据作为几何块进行正交匹配追踪,通过几何块的方法可使数据的完整性得到保证,最后结合小波与块字典对坡度数据进行混合插值。稀疏混合估计法通过同时计算L1和L2范式,充分考虑到坡度数据的规律性变化,结合多方向小波变换在不同方向对坡度数据进行混合插值,保证了重构数据的完整性和结果的准确率。本研究以甘肃龙泉梯田区DEM为研究对象,由无人机航空摄影测量方法获取高分辨率高程数据,并从中提取坡度。利用改进的POCS算法和稀疏混合估计算法针对DEM坡度数据进行超分辨率重构,并与最近邻法、双线性插值法、三次卷积插值法比较与评估。结果表明本研究方法在空间分布和误差方面上均优于其它方法。在本研究的两个算法结果比较中,稀疏混合估计法的结果要优于改进的POCS算法。
[Abstract]:Slope (Slope) is not only an important parameter of topographic features, but also an important evaluation index of geomorphologic change. Slopes on a regional scale are usually calculated by DEM (Digital Elevation Model). Because of the difficulty of obtaining high-resolution DEM data, and the loss of high-frequency and low-frequency parts of slope data at low resolution, Therefore, it is one of the ways to obtain high precision slope data by super-resolution reconstruction (downscaling change). In this study, the super-resolution reconstruction of slope data is studied based on the improved POCS (Projections Onto Convex Sets) algorithm and the sparse mixed estimation method. The main contents are as follows: (1) Super-resolution reconstruction of slope data is carried out through the improved POCS algorithm. In this algorithm, we first judge the multiple groups of DEM data read in, determine the approximate range of the data values according to the variance results of each data, and eliminate the data which is different from the average range, so as to obtain a better reconstruction effect. The method of bilinear interpolation is used to construct the reference data to make the results smoother and truer, and then the reference data are corrected. The correction process is to detect the edge of the slope data according to the large change of terrace slope. After obtaining the detection results and modifying them to the reference data. (2) the slope data is reconstructed by super-resolution based on sparse mixed estimation method. Firstly, wavelet transform is carried out on the slope data in approximate, horizontal, vertical and 45 掳directions. This step can fully extract the feature of slope data. Secondly, the dictionary of slope data block and the orthogonal matching tracing of the block are established. By constructing the sparse redundant block dictionary of slope data, the characteristic of slope data is induced, and the data is used as geometric block for orthogonal matching tracing. The integrity of the data can be guaranteed by geometric block method. Finally, the slope data is interpolated with wavelet and block dictionary. By calculating L _ 1 and L _ 2 normal forms at the same time, the sparse mixed estimation method takes full account of the regular variation of slope data, and combines multi-directional wavelet transform to interpolate gradient data in different directions. The integrity of the reconstructed data and the accuracy of the results are ensured. In this study, high resolution elevation data were obtained from aerial photogrammetry of unmanned aerial vehicle (UAV) and slope was extracted from DEM in Longquan terraced area of Gansu province. The improved POCS algorithm and sparse mixed estimation algorithm are used to reconstruct the DEM slope data in super-resolution, and compared with the nearest neighbor method, bilinear interpolation method and cubic convolution interpolation method. The results show that this method is superior to other methods in spatial distribution and error. In the comparison of the results of the two algorithms, the result of sparse mixed estimation is better than that of the improved POCS algorithm.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P208

【参考文献】