Hyperspectral Unmixing-Based Anomaly Detection

Abstract

Research supporting improved anomaly detection performance benefits a wide range of technical applications. Thus, the definition of anomalies and the subsequent means to detect them are wide-ranging. This treatment presents an overview of the development of an anomaly detection approach based on spectral signatures obtained with hyperspectral unmixing. Anomaly Detection is a binary classification that does not require prior information about the anomaly. For example, an anomaly detector applied to hyperspectral imaging (HSI) would take a hyperspectral image with hundreds of channels as an input and output a two-dimensional image map of pixel intensities based on a threshold procedure applied to the probability of that pixel being an anomaly. There have been many advancements in the field of HSI Anomaly Detection. Our ensemble method algorithm, presented here, addresses some of the shortcomings of current state-of-the-art techniques. We present details about the extracted end-members and use them for effective anomaly detection. Our current ensemble method opens the path for future machine-learning processes. We evaluated our method on multiple datasets and reported the F1-macro score. We suggest that the Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve should not be used in Hyperspectral anomaly detection as an evaluation metric.

Publication Title

Proceedings of SPIE - The International Society for Optical Engineering

Link to Full Text

Share

COinS