Hyperspectral Anomaly Detection based on Frequency Analysis of Repeated Spatial Patterns

Hyperspectral anomaly detection has brought significant attention to researchers during the last decade. In the remote sensing domain, it has lots of applications like environmental monitoring, mine source detection, human safety and, etc. Different methods have been proposed to overcome the challenges of anomaly detection, which mainly are a highly accurate detection rate and also having low false alarm rates. Background complexity of the hyperspectral data causes false detection of pixels as anomalies, and several statistical and sparse based methods try to solve it. However, the computation burden of the statistical and sparse based methods make them unpractical. Also, they often do not utilize the spatial information of hyperspectral data for increasing the detection accuracy. In this paper, we proposed a frequency analysis based approach that suppresses the repeatable patterns of the scene which mainly represent the background and considers the remainder regions as anomalies. Owe to the Fourier analysis advantage, it fast and accurate detects anomalous regions. Two well-known data sets have been chosen to challenge the potency of the proposed method, and the results are compared with some state-of-the-art methods. The visual and quantitative results confirm the supremacy of the proposed method to competitors in computation complexity and detection accuracy point of view.