Translator Disclaimer
8 June 2011 Spectral analysis algorithm for material detection from multispectral imagery
Author Affiliations +
Material detection from multi-spectral imagery is critical to numerous geospatial applications. However, given the limited number of channels from various air and space-borne imaging sensors, coupled with varying illumination conditions, material-specific detection rules tend to generate large numbers of false positives. This paper will describe a novel approach that uses various band ratios (for example, [Blue + Green]/Red) to identify targets-of-interest, regardless of the illumination conditions and position of the sensor relative to the target. The approach uses a physics-based spectral model to estimate the observed channel-weighted radiance based on solar irradiance, atmospheric transmission, reflectivity of the target-of-interest and the spectral weighting functions of the sensor's channels. The observed channelweighted radiance is then converted to the expected channel pixel value by the channel-specific conversion factor. With each channel's pixel values estimated, the algorithm goes through a process to find which band ratio values show the least amount of variance, despite varying irradiance spectra and atmospheric absorption. The band ratios with the least amount of variance are then used to identify the target-of-interest in an image file. To determine the expected false alarm rate, the same band ratios are evaluated against a library of background materials using the same calculation method for determining the target-of-interest's channel pixel values. Testing of this approach against ground-truth imagery, with as few as four channels, has shown a high rate of success in identifying targets-of-interest, while maintaining low false alarm rates.
© (2011) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Joseph K. Racine "Spectral analysis algorithm for material detection from multispectral imagery", Proc. SPIE 8040, Active and Passive Signatures II, 80400H (8 June 2011);

Back to Top