<title>Effect of light level and photon noise on hyperspectral target detection performance</title>

Abstract

The performance of hyperspectral target detection is known to vary with the light level. To characterize this variation, the effect of synthetically varying the amount of light is studied, starting with real images. Camera calibration coefficients are divided out to obtain the number of detector photoelectrons corresponding to each radiance sample. It is then straightforward to scale down the photoelectron numbers to represent a camera with lower light throughput and introduce additional noise by drawing corresponding Poisson-distributed photoelectron values. To first order, the resulting data may alternatively be considered to represent the increase in noise expected from a reduction in scene illumination. Images are generated with reduction of the light level down to a fraction of a percent of the original image. Sample spectra are shown after correction to photoelectron count, and it is argued that this representation of the data is useful because it permits direct estimation of signal to noise ratio. Spectral anomaly detection is applied to original and modified images, and the variation of false alarm rate with reduction factor is characterized. The results indicate that photon noise is the dominating cause of false alarms. There is no clear sign of false alarm contributions from the actual background clutter down to the data limit of false alarm probability, about 10-6 per pixel.© (2007) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.