Change detection using synthetic hyperspectral imagery

In the best of circumstances, change detection (CD) is accomplished using measurements from the same instrument and under similar collection circumstances. Complications in the CD process arise when the variability in the collection process is not minimized. Variations between collected images and a lack of precise corresponding ground truth make accurate evaluation of a given CD method imprecise at best. This work leverages synthetic hyperspectral imagery, with known ground truth to include primary and tertiary materials, to investigate the use of common CD algorithms for the hyperspectral CD problem. Specifically, we use synthetic hyperspectral images with different spatial resolutions acquired at different altitudes, thus exhibiting different atmospheric affects. The importance of this work is in definition of a CD taxonomy and using that taxonomy for the accurate evaluation of several CD methods. Results are presented using receiver operating characteristic (ROC) curves and the area under the ROC curve, indicating that, under mildly varying imaging conditions, principal component analysis-based CD outperforms simple image differencing and correlation coefficient-based CD methods.