A Context-Adaptive Hyperspectral Sensor and Perception Management Architecture for Airborne Anomaly Detection

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Airborne hyperspectral sensors capture extensive spectral data, enabling anomaly detection when target signatures are unknown.
• Real-world applications demand robust and adaptable anomaly detection, yet current research often uses limited datasets.
• Existing anomaly detection methods struggle with complex, dynamic environments, impacting reliability.

Purpose of the Study:

• To introduce a context-adaptive hyperspectral sensor and perception management (hSPM) architecture.
• To address the limitations of current anomaly detection methods in complex environments.
• To provide a large-scale, diverse airborne hyperspectral dataset for research.

Main Methods:

• Developed a novel hSPM architecture integrating sensor context extraction, band selection, and detector management.
• Evaluated the hSPM architecture on a new, large-scale airborne hyperspectral dataset (1100+ samples, 2 environments).
• Performed comparative experiments against state-of-the-art anomaly detection techniques.

Main Results:

• The hSPM architecture demonstrated superior detection accuracy and processing speed compared to conventional methods.
• Anomaly detection performance improved by 28-204% under varying conditions.
• Computation time was reduced by 70-99% using the hSPM architecture.

Conclusions:

• Adaptive sensor processing architectures like hSPM are crucial for robust airborne hyperspectral anomaly detection.
• Large, openly available datasets are essential for advancing the field.
• The proposed hSPM system offers significant improvements in accuracy and efficiency for real-world applications.