基于独立成分分析的高光谱变化检测

Hyperspectral change detection based on independent component analysis

现有的变化检测方法并未充分研究变化前后高光谱遥感影像端元的联系,不能准确地分析地物变化信息。本文提出了一种基于独立成分分析的高光谱遥感影像变化检测方法,对差值影像进行基于偏斜度的独立成分分析,在不同组分图中分别显示单一地物的变化情况,进而提取变化信息。实验表明,该方法能够在获得较高检测率的同时维持较低的误检率,检测效果优于传统方法。

Change detection is the process of analyzing changes of surface features with multi-temporal remote sensing imagery of the same area. Hyperspectral remote sensing images contain abundant spectral information for accurate change detection, which, regrettably, is not fully taken into account by existing approaches. In this paper, a hyperspectral change detection method based on Independent Component Analysis （ICA） is investigated. The difference image is analyzed by skew-based ICA. The change of a single feature can be obtained and then the change is extracted from each abundance image. Experiment results dem- onstrate that the ICA-based hyperspectral change detection performs better than other traditional methods with a high detection rate and a low false detection rate.