Unsupervised Anomaly Detection in Hyperspectral Imaging: Integrating Tensor Robust Principal Component Analysis With Autoencoding Adversarial Networks

Abstract

Hyperspectral (HS) image analysis has gained significant attention due to its ability to capture detailed spectral information across hundreds of bands, making it useful for environmental monitoring and mineral exploration applications. However, detecting anomalies in HS images, especially in complex scenes, remains challenging. This paper proposes a novel approach for robust anomaly detection by integrating tensor robust principal component analysis (TRPCA) with autoencoding adversarial networks (AEAN). Our method utilizes the AEAN model to learn a nonlinear low-dimensional representation of the spectral characteristics of background regions, which is then incorporated into the TRPCA framework. The TRPCA is further enhanced by incorporating prior knowledge of the sparsity of anomalous regions, enabling more accurate separation of background and anomaly components. This integration, achieved through a plug-and-play alternating direction method of multipliers (PnP-ADMM), significantly improves detection accuracy and robustness. Experimental results on benchmark datasets widely used for HS anomaly detection confirm that the proposed method consistently outperforms conventional techniques, achieving superior area-under-the-curve (AUC) scores across diverse and complex scenes. By leveraging both nonlinear modeling of background characteristics and sparsity-based anomaly separation, this research provides a more accurate and robust solution for HS anomaly detection, highlighting its potential for practical applications in remote sensing.