Object detection by using "whitening/dewhitening" to transform target signatures in multitemporal hyperspectral and multispectral imagery

Abstract

Changes in atmosphere, ground conditions, scene temperature, solar illumination, and sensor response can significantly affect the detected multispectral and hyperspectral data. Using uncorrected spectral target signatures in spectral matched filter searches therefore results in target detection with concomitant high false-alarm rates due to changes in multispectral and hyperspectral images. This letter introduces the use of the whitening/dewhitening (WD) transform to help correct target spectral signatures under varying conditions. An important feature of this transform is that it does not require subpixel registration between images collected at two distinct times. The transform was tested on images taken from two very different data collects using different sensors, targets, and backgrounds. In one dataset, the transform was applied to hyperspectral images taken from airborne longwave infrared sensor binned to 30 bands and the other data collect used images of a variety of tanks, trucks, calibration panels that were collected using bore-sighted broadband visible, shortwave infrared, midwave infrared, and longwave infrared staring array sensors. Target spectral signatures were transformed using imagery of spatially overlapping regions from datasets collected at different times and processed using the whitening and then dewhitening transform (inverse of a whitening transform). Use of the WD transform yielded a large target-to-clutter ratio (TCR) and was compared to the TCR derived from other transforms that approximated the cross-covariance matrix. In addition, the WD-transformed signatures applied in a matched filter search found targets (some concealed behind vegetative foliage or underneath camouflage) with low false-alarm rates as shown in a receiver operator characteristic curve. This letter demonstrates that the WD transform enhances searches for concealed targets in multisensor and hyperspectral data.