Temporal-Spectral Detection in Long-Wave IR Hyperspectral Imagery

Ground-based staring hyperspectral chemical detectors allow for repeated measurements through time with near-perfect image registration. The problem with standard spectral-based hyperspectral detection algorithms is that they do not make effective use of this temporal information. In this paper, we develop new temporal-spectral detection algorithms, and show that significant improvements in detection performance for staring geometry are achieved by making use of statistical and signal information obtained from previous samples. These new algorithms have the advantage that they limit detection to regions where both temporally and spectrally significant events have occurred. We present the development of these algorithms and demonstrate the performance of both temporal-spectral and spectral-only detectors for detection of gaseous plumes using data from a passive long-wave IR hyperspectral sensor.