Anomaly Detection of Hyperspectral Imagery Using Modified Collaborative Representation

Abstract

Hyperspectral images in remote sensing systems with rich spatial and spectral information provide an opportunity for researchers to discover the world. Anomaly detection is one of the most interesting topics over the last two decades in hyperspectral imagery (HSI). In this letter, we propose a modified collaborative-representation-based with outlier removal anomaly detector (CRBORAD) for anomaly detection. We use both spectral and spatial information for detecting anomalies since that is more precise than using only spectral information. The proposed detector can adaptively estimate the background by its adjacent pixels within a sliding dual-window. We remove outlier pixels that are significantly different from majority of pixels, before estimating background pixels. It can lead us to precise detection of anomalies in subsequent stages. By subtracting the predicted background from the original HSI, the residual image is resulted and anomalies can be determined, finally. Kernel extension of the proposed approach is also presented. CRBORAD results on San Diego airport and the Rochester Institute of Technology data are illustrated using intuitive images, receiver operating characteristic curves, and area under curve values. The results are compared with four popular and previous methods and prove the superiority of the proposed CRBORAD method.