Fast Processing of Massive Hyperspectral Image Anomaly Detection Based on Cloud-Edge Collaboration
With the improvement of hyperspectral image resolution, existing anomaly detection algorithms find it challenging to quickly process large volumes of hyperspectral data while fully exploiting spectral information. The collaborative cloud-edge computing, as an emerging computing paradigm, aims to int...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| 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/11015979/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | With the improvement of hyperspectral image resolution, existing anomaly detection algorithms find it challenging to quickly process large volumes of hyperspectral data while fully exploiting spectral information. The collaborative cloud-edge computing, as an emerging computing paradigm, aims to integrate cloud and edge computing paradigm for more efficient data processing. However, existing algorithms cannot be directly deployed directly in the cloud-edge collaborative environment for rapid detection. To address this issue, this article designs a collaborative cloud-edge anomaly detection method. First, to enhance the anomaly detection capability and improve the overall detection accuracy, we propose a joint subspace constraint representation (JSCR) model. We also introduce a corresponding joint dictionary construction algorithm aimed at performing subspace learning. Then, different detection task nodes are deployed at the edge and the cloud based on the characteristics of cloud-edge computing paradigm. Furthermore, we propose a cloud-edge model solving algorithm. This algorithm reformulates the JSCR model into a new optimization problem involving a small amount of factorized data. It minimizes the large-scale data transmission between the cloud and the edge while reducing the data volume required for model solving at the cloud. The proposed method in this article efficiently leverages the computational resources of both the cloud and the edge to execute anomaly detection. Experimental results demonstrate that, compared to existing hyperspectral anomaly detection algorithms, the proposed algorithm provides more accurate detection results in less time. |
|---|---|
| ISSN: | 1939-1404 2151-1535 |