Improved Central Attention Network-Based Tensor RX for Hyperspectral Anomaly Detection
Recently, using spatial–spectral information for hyperspectral anomaly detection (AD) has received extensive attention. However, the test point and its neighborhood points are usually treated equally without highlighting the test point, which is unreasonable. In this paper, improved central attentio...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-11-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/14/22/5865 |
| _version_ | 1827643709212065792 |
|---|---|
| author | Lili Zhang Jiachen Ma Baohong Fu Fang Lin Yudan Sun Fengpin Wang |
| author_facet | Lili Zhang Jiachen Ma Baohong Fu Fang Lin Yudan Sun Fengpin Wang |
| author_sort | Lili Zhang |
| collection | DOAJ |
| description | Recently, using spatial–spectral information for hyperspectral anomaly detection (AD) has received extensive attention. However, the test point and its neighborhood points are usually treated equally without highlighting the test point, which is unreasonable. In this paper, improved central attention network-based tensor RX (ICAN-TRX) is designed to extract hyperspectral anomaly targets. The ICAN-TRX algorithm consists of two parts, ICAN and TRX. In ICAN, a test tensor block as a value tensor is first reconstructed by DBN to make the anomaly points more prominent. Then, in the reconstructed tensor block, the central tensor is used as a convolution kernel to perform convolution operation with its tensor block. The result tensor as a key tensor is transformed into a weight matrix. Finally, after the correlation operation between the value tensor and the weight matrix, the new test point is obtained. In ICAN, the spectral information of a test point is emphasized, and the spatial relationships between the test point and its neighborhood points reflect their similarities. TRX is used in the new HSI after ICAN, which allows more abundant spatial information to be used for AD. Five real hyperspectral datasets are selected to estimate the performance of the proposed ICAN-TRX algorithm. The detection results demonstrate that ICAN-TRX achieves superior performance compared with seven other AD algorithms. |
| first_indexed | 2024-03-09T18:01:43Z |
| format | Article |
| id | doaj.art-1fde7d1732f4471e886c5f2e6e8a3163 |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-09T18:01:43Z |
| publishDate | 2022-11-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-1fde7d1732f4471e886c5f2e6e8a31632023-11-24T09:51:41ZengMDPI AGRemote Sensing2072-42922022-11-011422586510.3390/rs14225865Improved Central Attention Network-Based Tensor RX for Hyperspectral Anomaly DetectionLili Zhang0Jiachen Ma1Baohong Fu2Fang Lin3Yudan Sun4Fengpin Wang5College of Mechanical and Electrical Engineering, Daqing Normal University, Daqing 163712, ChinaSchool of Computer Science and Technology, Heilongjiang University, Harbin 150080, ChinaCollege of Mechanical and Electrical Engineering, Daqing Normal University, Daqing 163712, ChinaCollege of Mechanical and Electrical Engineering, Daqing Normal University, Daqing 163712, ChinaCollege of Mechanical and Electrical Engineering, Daqing Normal University, Daqing 163712, ChinaCollege of Mechanical and Electrical Engineering, Daqing Normal University, Daqing 163712, ChinaRecently, using spatial–spectral information for hyperspectral anomaly detection (AD) has received extensive attention. However, the test point and its neighborhood points are usually treated equally without highlighting the test point, which is unreasonable. In this paper, improved central attention network-based tensor RX (ICAN-TRX) is designed to extract hyperspectral anomaly targets. The ICAN-TRX algorithm consists of two parts, ICAN and TRX. In ICAN, a test tensor block as a value tensor is first reconstructed by DBN to make the anomaly points more prominent. Then, in the reconstructed tensor block, the central tensor is used as a convolution kernel to perform convolution operation with its tensor block. The result tensor as a key tensor is transformed into a weight matrix. Finally, after the correlation operation between the value tensor and the weight matrix, the new test point is obtained. In ICAN, the spectral information of a test point is emphasized, and the spatial relationships between the test point and its neighborhood points reflect their similarities. TRX is used in the new HSI after ICAN, which allows more abundant spatial information to be used for AD. Five real hyperspectral datasets are selected to estimate the performance of the proposed ICAN-TRX algorithm. The detection results demonstrate that ICAN-TRX achieves superior performance compared with seven other AD algorithms.https://www.mdpi.com/2072-4292/14/22/5865anomaly detectioncentral attentiontensorhyperspectral image |
| spellingShingle | Lili Zhang Jiachen Ma Baohong Fu Fang Lin Yudan Sun Fengpin Wang Improved Central Attention Network-Based Tensor RX for Hyperspectral Anomaly Detection Remote Sensing anomaly detection central attention tensor hyperspectral image |
| title | Improved Central Attention Network-Based Tensor RX for Hyperspectral Anomaly Detection |
| title_full | Improved Central Attention Network-Based Tensor RX for Hyperspectral Anomaly Detection |
| title_fullStr | Improved Central Attention Network-Based Tensor RX for Hyperspectral Anomaly Detection |
| title_full_unstemmed | Improved Central Attention Network-Based Tensor RX for Hyperspectral Anomaly Detection |
| title_short | Improved Central Attention Network-Based Tensor RX for Hyperspectral Anomaly Detection |
| title_sort | improved central attention network based tensor rx for hyperspectral anomaly detection |
| topic | anomaly detection central attention tensor hyperspectral image |
| url | https://www.mdpi.com/2072-4292/14/22/5865 |
| work_keys_str_mv | AT lilizhang improvedcentralattentionnetworkbasedtensorrxforhyperspectralanomalydetection AT jiachenma improvedcentralattentionnetworkbasedtensorrxforhyperspectralanomalydetection AT baohongfu improvedcentralattentionnetworkbasedtensorrxforhyperspectralanomalydetection AT fanglin improvedcentralattentionnetworkbasedtensorrxforhyperspectralanomalydetection AT yudansun improvedcentralattentionnetworkbasedtensorrxforhyperspectralanomalydetection AT fengpinwang improvedcentralattentionnetworkbasedtensorrxforhyperspectralanomalydetection |