论文部分内容阅读
为了进一步提高基于独立分量分析ICA(Independent Component Analysis)的遥感图像变化检测精确度,更好地实现地表覆盖的动态监测,将多尺度几何分析和核独立分量分析KICA(Kernel Independent Component Analysis)相结合应用于遥感图像的地表覆盖变化检测。首先利用Contourlet变换、复Contourlet变换CCT(Complex Contourlet Transform)、非下采样Contourlet变换NSCT(Nonsubsampled Contourlet Transform)等多尺度几何分析对土地遥感图像进行多尺度分解;然后对分解后的数据进行核独立分量分析,通过核函数将数据映射到高维特征空间中,再在该空间中用ICA方法分离出互相独立的分量;最后将分离后的地表变化分量转化为图像分量,再采用最大类间方差法对变化图像进行分割,实现地表覆盖的变化检测。给出了本文方法和近年来提出的基于ICA、基于KICA、基于小波变换和ICA等变化检测方法的实验结果,并进行了分析和定量比较。实验结果表明,基于多尺度几何分析和KICA的变化检测方法能更好地分离出遥感图像的变化信息,其中基于NSCT和KICA方法的错判和漏检误差最小,且在边缘细节方面处理得更好,而基于Contourlet变换和KICA方法的检测效率相对较高。
In order to further improve the detection accuracy of remote sensing images based on Independent Component Analysis (ICA), and to achieve better dynamic monitoring of surface coverage, multi-scale geometric analysis and kernel independent component analysis (KICA) are combined Land cover change detection applied to remote sensing images. Firstly, multiscale decomposition of land remote sensing image is carried out by means of multi-scale geometric analysis such as Contourlet transform, Complex Contourlet Transform (CCT) and Nonsubsampled Contourlet Transform (NSCT). Then the kernel independent component Then the data are mapped into the high-dimensional feature space by kernel function, and then the ICA method is used to separate the independent components in the space. Finally, the separated surface variation components are converted into image components, and the maximum inter-class variance Segmentation of changes in the image, to achieve changes in surface coverage detection. The experimental results based on ICA, KICA-based, wavelet transform-based and ICA-based detection methods proposed in this paper and in recent years are given and analyzed and quantitatively compared. The experimental results show that the change detection method based on multi-scale geometric analysis and KICA can better separate the change information of remote sensing images. The errors of misjudgment and missed detection based on NSCT and KICA methods are the smallest, and the edge details are handled more However, the detection efficiency based on Contourlet transform and KICA method is relatively high.