Radiometric Normalization Of Remote Sensing Image Based On Kernel Canonical Correlation Analysis

In the target detection and change monitoring of remote sensing data,the processing and analysis of multi-sensors,large-area,and time-series remote sensing data are the primary tasks.Among them,radiometric differences and the natural changes of vegetation have brought serious disturbances to the monitoring of changes.How to eliminate the interference caused by this kind of noise is one of the urgent problems to be solved in the remote data analysis.At present,a large number of researchers have proposed many rapid,simple,and effective radiometric normalization methods.This dissertation summarizes these methods,and finds that there are still some deficiencies and limitations in the existing methods,mainly reflected in the following aspects: most of the methods are based on linear assumptions,it is only an approximate linear relationship;there are strict constraints on the target image,such as the time phase of the target image and the reference image are must very similar,which greatly limits the application of relative radiometric normalization;most of the researches focused on the selection of invariant points,but little attention is paid to the fitting process.;only the normalization of pseudo-invariant points was taken into account without considering the features whose spectral reflectance changes regularly or expectably over time;the research on radiometric normalization based on nonlinear methods is very limited and cannot be applied to practical projects.Against the above problems,the dissertation introduced a multivariate statistical analysis method-kernel canonical correlation analysis in the field of radiometric normalization.It is compared and analyzed with the existing methods,and finally applied to the cloud detection.The main research directions and contributions of this paper are listed as follows:(1)The improved canonical correlation analysis based method is proposed.This paper aims to eliminate the fitting error caused by the “noise points” in the pseudo-invariant points extracted by canonical correlation analysis method,proposes a fitting method based on Ransac robust regression.This method improves the accuracy of gain and bias,eliminates the overall radiometric difference caused by external factors between images,and makes radiometric normalized results more accurate.(2)The dissertation proposes a kCCA based method for radiometric normalization.This method uses the multi-analysis capability of the kCCA to extract the nonlinear relationship between the multi-temporal images,and normalizes the target image.The kCCA-based normalization can preserve more similarity and better correlation between an image-pair and effectively avoid the color error propagation.The proposed method not only builds the common scale or reference to make the radiometric consistency among GF-1 image sequences,but also highlights the interesting spectral changes while eliminates less interesting spectral changes.Our method enables the application of GF-1 data for change detection,land-use,land-cover change detection etc.(3)The dissertation proposes a kCCA based method for cloud detection.This paper proposes an automatic and fast cloud detection algorithm combining radiometric normalization with change detection for multi-temporal images,as an application of the kCCA based radiometric normalization.The method uses the extracted invariant points to perform radiometric normalization,and uses extracted changes pixels to do cloud detection.Finally,the cloud pixels are patched by the normalized result.The experimental results show that this method can be effectively applied to the cloud detection and cloud compensation,and accurate cloud mask and cloud-remove products with good radiometric consistency can be obtained.This is also a significant application of the kCCA methods.