Assessment of Spatial–Spectral Feature-Level Fusion for Hyperspectral Target Detection

Abstract

In this work, we assess the detection and classification of specially constructed targets in coincident airborne hyperspectral imagery (HSI) and high spatial resolution panchromatic imagery (HRI) in spectral, spatial, and joint spatial-spectral feature spaces. The target discrimination powers of the data-level and feature-level fusion of HSI and HRI are also directly compared in the spatial-spectral context using airborne imagery collected explicitly for this research. We show that in the case of Bobcat 2013 imagery, feature-level fusion of the HSI spectrum with spatial features derived from the coincident HRI data consistently results in fewer false alarms on the scene background as well as fewer misclassifications among the tested targets. Furthermore, this approach also outperforms schemes in which data-level fusion of the HSI and HRI imagery is performed prior to extracting spatial-spectral features.Index Terms-Hyperspectral imagery (HSI), image fusion, material identification, pansharpening, spatial-spectral feature extraction, target identification.

On the effects of spatial and spectral resolution on spatial-spectral target detection in SHARE 2012 and Bobcat 2013 hyperspectral imagery

“…ARCHER's imaging spectrometer has 52 bands spanning 504.6nm to 1100nm and produces uncalibrated spectral data for 504 spatial samples. We utilized the same subset of four images from [4] but processed them with an improved scene-based nonuniformity correction algorithm and performed ELM atmospheric compensation. 13 The native GSD of this imagery was near 40cm, but we also produced 100cm geo-corrected imagery for direct comparisons with the ProSpecTIR-VS imagery that was taken at 100cm.…”

“…[4][5][6] The spatial feature vectors produced about each spatial pixel from each spectral image band can be concatenated, forming a spatial-spectral feature vector or pseudospectrum that simultaneously captures spectral, shape, and textural attributes of a pixel. This concept originates from S-DAISY, 5 an extension to the local image descriptor DAISY.…”

“…In [4], descriptors robust to spatial transformations, namely, Hu moments, 8 Suk moments, 9 and pseudoZernike moments (PZMs) 10 along with the aforementioned S-DAISY were applied to HSI, coincident panchromatic imagery, and pansharpened HSI. There, it was determined that S-DAISY and PZMs produced significantly lower target detection false alarm rates than the other spatial descriptors.…”

“…We adopt the robust average number of confusers at full association (ANCFA) metric first presented in [4]. To compute the robust ANCFA, we first produce an ACE detection statistic image ACE t for each signature s t extracted from the F 0 training feature image against the entire test feature image F 1 .…”

“…Previous work with the Bobcat 2013 data set showed that spatial-spectral exploitation of VNIR HSI data led to better target detection and discrimination but did not consider the possible benefits of the SWIR portion of the spectrum. 3,4 In this work, we evaluate the tradeoffs in spatial and spectral resolution and spectral coverage between the two imagery collections for a common set of targets in terms of their effects on spatial-spectral target detection performance. The remainder of this paper is laid out as follows: in Section 2, we review the exploitation framework through which the HSI imagery prepared in Section 3 will be processed.…”