基于主成分变换的多光谱遥感数据的有监督可视化方法

【目的】遥感数据可视化是决定遥感数据解译质量高低的一个关键环节,目视解译是目前生产实践中遥感技术应用的一个重要方式。探讨一种基于学习样本生成可视化特征,使得可视化与目视解译应用能够有机结合,进而实现有监督可视化。【方法】采用主成分变换回归重构方法重构了针对森林资源的植被监测模型,得到3个新特征,依次赋予红绿蓝3种颜色以生成假彩色图像来实现信息可视化。【结果】基于学习样本的主成分变换回归重构方法与主成分变换法、原始近红外-短波红外-红波段组合相比,重构后的假彩色图像在一定程度上得到了改善,有利于目视解译。主成分变换回归重构后的分类精度比未重构的分类精度提高了6.20%,比原始波段组合的分类精度提高了7.82%。重构前后的一元和多元离差平方和(离差阵)构成变化分析说明,重构后类内差异缩小、类间差异增大。【结论】定量验证认为,主成分变换回归重构的可分性比未重构的主成分变换和原始波段组合的可分性都要好。

Supervised visualization of multispectral remotely sensed data based on principal component transformation

【Objective】Visualization of remotely sensed data is a key step to determine the interpretation quality of remotely sensed data. Visual interpretation is an important way of applying remotely sensed technology in current production practice. To exploring a method for generating visual features which based on learning samples, so that visual and visual interpreting applications can be organically combined to achieve supervised visualization. 【Method】Using the newly proposed regression reconstruction of principal component transform coefficients(RRPCTC)to constructed the model,three new features are obtained,which are respectively assigned red,green and blue three colors in order to generate false color images for visualization.【Result】The image produced by RRPCTC is improved to a certain extent compares with the ordinary principal component transformation(OPCT)method and the original near infrared-short wave infrared-red 3-band combination. It is helpful for visual interpretation. The classification accuracy of RRPCTC is improved by 6.20% compared with that of the OPCT and by 7.82% compared with that of the original 3-band combination. The changes of sums of squares of deviations of the first 3 PCs respectively and the deviation matrix before and later reconstruction show that the reconstruction makes the differences smaller inner classes and larger between classes.【Conclusion】All these quantitatively prove that the separability of the RRPCTC is better than that of OPCT and that of original 3-band combination.