Anomaly detection using an adaptive algorithm for estimating mixtures of backgrounds in hyperspectral images

Anomaly detection in hyperspectral data has been considered for various applications. The main purpose of anomaly detection is to detect pixel vectors (i.e. spectral vectors) whose spectra differ significantly from the background spectra. In anomaly detection, no prior knowledge about the target is assumed. In this paper we will present a new method for anomaly detection based on the SRX (Segmented RX) algorithm, with an emphasis on the edges between the segments. This method incorporates an adaptive algorithm with fast convergence which we developed for estimating the mixing coefficients of adjacent segments to fit the spectra of edge pixels. Achieving it allows us to reconstruct its mean vector and its covariance matrix, and operate the RX algorithm locally. The developed algorithm is a fusion and improvement of two algorithms (Steepest Descent and Newton's Method); it combines the benefits of each method while eliminating their drawbacks, so its convergence is fast and stable.