Automatic target detection and recognition in multiband imagery: a unified ML detection and estimation approach

Area of Science:

• Remote Sensing
• Signal Processing
• Computer Vision

Background:

• Natural clutter (vegetation) and man-made objects exhibit distinct spectral characteristics.
• Multispectral and hyperspectral sensors capture detailed spectral-spatial information.
• Detecting low-contrast targets in clutter is challenging due to spectral similarities.

Purpose of the Study:

• To develop a unified framework for automatic target detection and recognition (ATDR).
• To generalize existing data models for target detection in clutter.
• To evaluate and compare the performance of various multiband detectors.

Main Methods:

• Formulating a generalized hypothesis test based on spectral signatures.
• Partitioning spectral bands into target-dominant and clutter-dominant groups.
• Utilizing a maximum likelihood ratio approach for detection.
• Evaluating performance by analyzing SNR gain and target separability.

Main Results:

• A unified framework for ATDR is presented, generalizing previous approaches.
• The framework allows for performance evaluation and comparison of multiband detectors.
• Incremental gains in SNR and separability were studied using target-feature and clutter-reference bands.
• Key parameters influencing SNR and separability gains were identified.

Conclusions:

• The proposed unified framework provides a robust approach to ATDR in cluttered scenes.
• The framework facilitates the optimization of sensor band selection for improved detection and recognition.
• Understanding the influence of specific bands enhances the effectiveness of multispectral and hyperspectral target detection systems.