Adaptive Subspace Target Detection in Hyperspectral Imagery

* Corresponding author. Abstract –Adaptive subspace detectors are widely used for low probability and anomaly detection. The complex remote sensing conditions in which hyperspectral imagery is obtained make the detector performance evaluation a non-trivial task. Many of the detector design parameters can only be studied empirically for their effects on detection performance. In this paper, hyperspectral images are generated using target and background endmember spectra based on the linear mixing model, where additive Gaussian noise is also added to the mixture model. An adaptive subspace detector is then applied to detect target pixels and the performance of the detector is investigated by varying the target abundance, noise variance, distribution of the target abundance, and target subspace dimension. The results show that the actual performance of the detector is highly dependent upon all of the four design parameters.