遥感反演场数据会由于云雾、地物的遮挡,传感器性能等原因造成部分区域数据的缺失而影响遥感反演场数据的应用。矩阵填充理论针对低秩矩阵,利用矩阵的低秩性,即数据的高相关性,可以高精度地对低秩矩阵中的缺值数值进行恢复,其中矩阵填充理论中的SVT(Singular Value Thresholding)算法可以对矩阵中缺失数值进行快速、高精度的估计,应用广泛。本文应用矩阵填充理论的SVT算法,以缺值点为中心,方差最小作为窗口尺度选择的标准,这样可以保证区域数据的高相关性,建立局部窗口,对窗口进行SVT算法填充。本文也针对相同缺值区域进行了距离反比加权插值、Kriging插值法插值和整体SVT算法插值,整体SVT算法插值即并未对缺值点进行相关性窗口判断,而是直接对整个区域进行SVT填充。并对这几种方法的精度进行比较,得到局部SVT算法的精度相比整体SVT算法和距离反比加权插值算法的精度要高,与Kriging算法相比,其精度变化趋势相似,在锋面区域局部SVT算法精度比Kriging方法要高。

Due to shading of clouds and objects on the ground and due to the performance of sensors, the part region of field inversion data by remote sensing may be incomplete, which makes it harmful to use these data for further study. The way to recover the incomplete field inversion data accurately and quickly seems to be significant. Matrix completion (MC) has been proposed in recent years, which is mainly used for low rank matrix. Because of the quality of low rank, the data of the matrix has high correlation, so the matrix can be high accurately recovered with MC. SVT (Singular Value Thresholding) algorithm is one method of MC, which could recover missing values in the matrix rapidly and accurately. In this paper, we introduced the SVT algorithm for matrix completion and we used this algorithm to complete the missing data points which were selected in different regions. Making the missing data points be the center of a region of square and the size of the square is selected by the criteria of the smallest variance. Use SVT algorithm to complete this square region and name this method local SVT algorithm (LSVT). Comparing with the SVT algorithm for the whole region data that is named WSVT which is based on the whole experiment area, inversing distance weighting method (IDW) and Kriging method respectively, we conclude that the precision for LSVT is higher than WSVT and IDW method. Also, the precision changing trend for LSVT is similar with Kriging method and the precision for LSVT is higher than Kriging method in sea front region.