Optimizing Change Detection for Planetary Remote Sensing Datasets

Abstract

The fast-growing Remote Sensing dataset for Mars has challenged scientists in the community to make full use of the imagery collected from spacecraft. Cameras aboard long-lived orbiter missions capturing single-band, high-resolution images provide a special opportunity to monitor surface and atmospheric changes associated with geologic, weather and climate processes. A potential method for detecting changes is to employ principle component analysis (PCA) - an existing image processing method normally applied to multi-spectral imagery. When PCA is applied to a time-series stack of single-band images of the same location, subtle changes in pixel brightness are enhanced, and are represented as a series of transformed image components. PCA of a time-series image stack sorts these variations in the dataset according to their amplitudes. The most significant pixel variations are found in the lower principle components (PCs) while more subtle changes in pixel brightness are revealed in higher PCs. This method makes visual identification of subtle, ongoing geologic processes easier as is shown in a series of HiRISE images containing recurring slope lineae (RSL). As a final step, the change detection method proposed herein can estimate the likelihood of each of the original images containing the feature found in the PCA components. This is performed via a PCA-to-image coordinate rotation using the Eigen matrix calculated from the original PCA analysis. This proposed method is called “Change Detection via PCA of Stacked Time-series” (CDPCAST).