Spectral-Spatial Feature Fusion for Hyperspectral Anomaly Detection.

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Hyperspectral anomaly detection identifies unusual patterns in hyperspectral data.
• Existing spectral-spatial methods often fail to leverage complementary spectral and spatial information, leading to high false alarm rates.

Purpose of the Study:

• To develop a novel spectral-spatial information fusion (SSIF) method for hyperspectral anomaly detection.
• To address the limitations of cascaded spectral-spatial approaches and reduce false alarms.

Main Methods:

• An isolation forest was used to generate a spectral anomaly map, with object-level features extracted via entropy rate segmentation.
• A local spatial saliency detection scheme was employed to derive spatial anomaly results.
• Spectral and spatial anomaly scores were fused and refined using domain transform recursive filtering.

Main Results:

• The proposed SSIF method demonstrated superior performance compared to state-of-the-art techniques.
• Experiments on diverse hyperspectral datasets (ocean, airport scenes) validated the effectiveness of the SSIF approach.
• The fusion strategy successfully mitigated the high false alarm rates associated with previous methods.

Conclusions:

• The SSIF method offers a robust and effective solution for hyperspectral anomaly detection.
• Integrating spectral and spatial information synergistically enhances detection accuracy and reduces false positives.
• The approach shows significant promise for real-world applications in analyzing hyperspectral imagery.