Matrix Autoregressive Model for Hyperspectral Anomaly Detection

Abstract

For anomaly detection in hyperspectral imagery, the scene can be treated as a combination of the background and the anomalies. Once a pure background hyperspectral image (HSI) is obtained, the anomalies can be easily located. This paper detects the anomalies via a matrix autoregressive model (MARM) to reconstruct the background HSI. Specifically, some informative and discriminative bands are firstly selected and come into a new HSI with less bands. Secondly, the new HSI can be treated as a collection of profiles in the row direction. Based on this, the background can be regularly reconstructed via the MARM. The regressive model not only respects the original matrix structure in the row profiles, but also utilizes both the spatial and spectral correlation for the detection process. Finally, the classical Reed Xiaoli detector is applied to the difference cube between the band selected HSI and the HSI reconstructed by MARM, achieving a final detection map with higher accuracy. Experimental results and data analysis on four different sensors captured datasets with different resolutions have validated the effectiveness of the proposed method.

Recursive RX with Extended Multi-Attribute Profiles for Hyperspectral Anomaly Detection

“…A hyperspectral image (HSI) provides a rich source of spectral and spatial information about the materials in the scene [1], and it has been widely applied in many remote sensing areas [2][3][4], including classification [5][6][7][8][9], clustering [10][11][12], unmixing [13][14][15], image denoising [16,17], band selection [18,19], change detection [20], and target detection [21][22][23][24][25] or anomaly detection [26][27][28][29][30][31][32][33]. Among these applications, anomaly detection (AD) plays a significant role in military surveillance [34], agriculture [35], mineral exploration [36], environmental monitoring [34], maritime rescue [37], and so on [27,[38][39][40][41].…”