A Band-Selected and Regularized Network for Hyperspectral Anomaly Detection

Hyperspectral anomaly detection (HAD) is crucial for its ability to identify targets without prior knowledge. Most existing methods reconstruct the scene to isolate and suppress the background. In this article, we propose a band selection and regularization network (BSRegNet)—a novel network that integrates band selection (BS) and regularization. BS serves as a preprocessing step to reduce data volume and eliminate redundancy, thereby enhancing discrimination between background and anomalies. We also introduce a regularization term that minimizes the first-order derivatives of the reconstructed background, promoting spectral smoothness. While BS improves detection accuracy and reduces computational load, the regularization term enhances background reconstruction and anomaly localization. Extensive experiments on multiple datasets demonstrate that BSRegNet outperforms existing methods, validating the effectiveness of our approach. The code is released at https://github.com/rk-rkk/A-Band-Selected-and-Regularized-Network-for-Hyperspectral-Anomaly-Detection