Hyperspectral Anomaly Detection Using Dual-Branch Network Based on Frequency Domain Learning

Existing deep learning-based hyperspectral anomaly detection methods often overlook frequency domain features, hindering the ability to effectively distinguish between background and anomalies. Furthermore, many methods directly apply Mahalanobis distance to reconstructed hyperspectral image (HSI) f...

Full description

Saved in:
Bibliographic Details
Main Authors: Xiaoyi Wang, Jun Du, Qingxuan Lv, Peng Wang, Xianchao Zhang, Jian Cheng, Henry Leung
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11038935/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850085429003419648
author Xiaoyi Wang
Jun Du
Qingxuan Lv
Peng Wang
Xianchao Zhang
Jian Cheng
Henry Leung
author_facet Xiaoyi Wang
Jun Du
Qingxuan Lv
Peng Wang
Xianchao Zhang
Jian Cheng
Henry Leung
author_sort Xiaoyi Wang
collection DOAJ
description Existing deep learning-based hyperspectral anomaly detection methods often overlook frequency domain features, hindering the ability to effectively distinguish between background and anomalies. Furthermore, many methods directly apply Mahalanobis distance to reconstructed hyperspectral image (HSI) for detection, disregarding the structural features of the original HSI. This leads to insufficient representation of important information and ultimately limits detection performance. To resolve these challenges, this article presents a dual-branch network based on frequency domain learning. Using the Haar wavelet transform, the original HSI is divided into high- and low-frequency components. Based on the edge detailed properties of the high-frequency component and the smoothness of the low-frequency component, distinct network branches are designed for feature extraction, with the extracted features fused for HSI reconstruction. Notably, the detection process utilizes the outputs of the high- and low-frequency branches directly, rather than the reconstructed HSI. To fully leverage the key structural information in the original HSI, a Mahalanobis distance detection method incorporating the structural similarity index is proposed. By weighting the covariance matrix, the method enhances critical structural and spectral features, improving detection accuracy while effectively suppressing noise. Experiments on six datasets demonstrate the proposed method’s superiority and robustness over eight advanced hyperspectral anomaly detection methods.
format Article
id doaj-art-7791643cdc97474f92aa03868660e36c
institution DOAJ
issn 1939-1404
2151-1535
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
spelling doaj-art-7791643cdc97474f92aa03868660e36c2025-08-20T02:43:43ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing1939-14042151-15352025-01-0118157891580310.1109/JSTARS.2025.358057511038935Hyperspectral Anomaly Detection Using Dual-Branch Network Based on Frequency Domain LearningXiaoyi Wang0https://orcid.org/0000-0002-1010-0158Jun Du1https://orcid.org/0000-0001-8576-1041Qingxuan Lv2Peng Wang3https://orcid.org/0000-0002-3825-6365Xianchao Zhang4https://orcid.org/0000-0001-8925-8371Jian Cheng5Henry Leung6https://orcid.org/0000-0002-5984-107XJoint Laboratory of Spatial Intelligent Perception and Large Model Application, Nanjing University of Aeronautics and Astronautics, Nanjing, ChinaJoint Laboratory of Spatial Intelligent Perception and Large Model Application, Nanjing University of Aeronautics and Astronautics, Nanjing, ChinaJoint Laboratory of Spatial Intelligent Perception and Large Model Application, Nanjing University of Aeronautics and Astronautics, Nanjing, ChinaJoint Laboratory of Spatial Intelligent Perception and Large Model Application, Nanjing University of Aeronautics and Astronautics, Nanjing, ChinaProvincial Key Laboratory of Multimodal Perceiving and Intelligent Systems, Jiaxing University, Jiaxing, ChinaJoint Laboratory of Spatial Intelligent Perception and Large Model Application, Nanjing University of Aeronautics and Astronautics, Nanjing, ChinaDepartment of Electrical and Computer Engineering, University of Calgary, Calgary, CanadaExisting deep learning-based hyperspectral anomaly detection methods often overlook frequency domain features, hindering the ability to effectively distinguish between background and anomalies. Furthermore, many methods directly apply Mahalanobis distance to reconstructed hyperspectral image (HSI) for detection, disregarding the structural features of the original HSI. This leads to insufficient representation of important information and ultimately limits detection performance. To resolve these challenges, this article presents a dual-branch network based on frequency domain learning. Using the Haar wavelet transform, the original HSI is divided into high- and low-frequency components. Based on the edge detailed properties of the high-frequency component and the smoothness of the low-frequency component, distinct network branches are designed for feature extraction, with the extracted features fused for HSI reconstruction. Notably, the detection process utilizes the outputs of the high- and low-frequency branches directly, rather than the reconstructed HSI. To fully leverage the key structural information in the original HSI, a Mahalanobis distance detection method incorporating the structural similarity index is proposed. By weighting the covariance matrix, the method enhances critical structural and spectral features, improving detection accuracy while effectively suppressing noise. Experiments on six datasets demonstrate the proposed method’s superiority and robustness over eight advanced hyperspectral anomaly detection methods.https://ieeexplore.ieee.org/document/11038935/Frequency domain decompositionmultiscale convolutionstructural similarity index (SSIM)-weighted Mahalanobis distancetransformer
spellingShingle Xiaoyi Wang
Jun Du
Qingxuan Lv
Peng Wang
Xianchao Zhang
Jian Cheng
Henry Leung
Hyperspectral Anomaly Detection Using Dual-Branch Network Based on Frequency Domain Learning
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Frequency domain decomposition
multiscale convolution
structural similarity index (SSIM)-weighted Mahalanobis distance
transformer
title Hyperspectral Anomaly Detection Using Dual-Branch Network Based on Frequency Domain Learning
title_full Hyperspectral Anomaly Detection Using Dual-Branch Network Based on Frequency Domain Learning
title_fullStr Hyperspectral Anomaly Detection Using Dual-Branch Network Based on Frequency Domain Learning
title_full_unstemmed Hyperspectral Anomaly Detection Using Dual-Branch Network Based on Frequency Domain Learning
title_short Hyperspectral Anomaly Detection Using Dual-Branch Network Based on Frequency Domain Learning
title_sort hyperspectral anomaly detection using dual branch network based on frequency domain learning
topic Frequency domain decomposition
multiscale convolution
structural similarity index (SSIM)-weighted Mahalanobis distance
transformer
url https://ieeexplore.ieee.org/document/11038935/
work_keys_str_mv AT xiaoyiwang hyperspectralanomalydetectionusingdualbranchnetworkbasedonfrequencydomainlearning
AT jundu hyperspectralanomalydetectionusingdualbranchnetworkbasedonfrequencydomainlearning
AT qingxuanlv hyperspectralanomalydetectionusingdualbranchnetworkbasedonfrequencydomainlearning
AT pengwang hyperspectralanomalydetectionusingdualbranchnetworkbasedonfrequencydomainlearning
AT xianchaozhang hyperspectralanomalydetectionusingdualbranchnetworkbasedonfrequencydomainlearning
AT jiancheng hyperspectralanomalydetectionusingdualbranchnetworkbasedonfrequencydomainlearning
AT henryleung hyperspectralanomalydetectionusingdualbranchnetworkbasedonfrequencydomainlearning