Robust Ground Target Detection by SAR and IR Sensor Fusion Using Adaboost-Based Feature Selection.

Area of Science:

• Remote Sensing
• Image Processing
• Machine Learning

Background:

• Detecting long-range ground targets in noisy environments using synthetic aperture radar (SAR) or infrared (IR) images is challenging.
• SAR detectors yield high detection rates but also high false alarms due to background noise.
• IR detectors are sensitive to weather conditions and primarily detect hot targets.

Purpose of the Study:

• To propose a novel, unified target detection method by fusing SAR and IR images.
• To achieve a high detection rate and low false alarm rate for ground targets.
• To overcome limitations of existing methods that use separate algorithms for SAR and IR data.

Main Methods:

• A modified Boolean map visual theory (modBMVT) incorporating a median local average filter (MLAF) and asymmetric morphological closing filter (AMCF).
• Automatic registration of heterogeneous SAR and IR images using RANdom SAmple Region Consensus (RANSARC)-based homography optimization.
• Feature-selection based sensor fusion using Adaboost for final target detection.

Main Results:

• The proposed modBMVT effectively removes noise and detects extended targets.
• RANSARC-based registration ensures accurate alignment of SAR and IR images.
• Adaboost fusion with feature selection demonstrates robust target detection performance.

Conclusions:

• The developed framework offers a unified approach for SAR and IR target detection.
• The fusion method significantly improves detection accuracy and reduces false alarms.
• The approach is validated on a synthetic database, showing promising performance.