Parametric adaptive modeling and detection for hyperspectral imaging

Hyperspectral imaging (HSI) sensors can provide very fine spectral resolution that allows remote identification of ground objects smaller than a full pixel. Traditional approaches to the so-called subpixel target signal detection problem involve the estimation of the sample covariance matrix of the background from target-free training pixels. This entails a large training requirement and high complexity. In this paper, we investigate parametric adaptive modeling and detection for HSI applications. To deal with nonstationarity in the spectral dimension that is characteristic of HSI data, we introduce a sliding-window based time-varying (TV) autoregressive (AR) modeling and detection technique, by which the spectral data is sliced into overlapping subvectors for parameter estimation and signal whitening. Experimental results using real HSI data show that the proposed parametric technique outperforms conventional detection schemes, especially when the training size is small.