Small Foreign Object Debris Detection for Millimeter-Wave Radar Based on Power Spectrum Features.

Area of Science:

• Radar Systems Engineering
• Signal Processing
• Machine Learning Applications

Background:

• Foreign Object Debris (FOD) poses significant risks in various operational environments.
• Accurate detection of FOD is crucial for safety and operational efficiency.
• Traditional classification methods struggle with distinguishing FOD from ground clutter in radar signals.

Purpose of the Study:

• To develop an advanced FOD detection system using a novel machine learning approach.
• To enhance the classification performance of millimeter-wave radar systems for FOD identification.
• To reduce the false alarm rate in FOD detection systems.

Main Methods:

• Extraction of echo features from FOD and ground clutter in the power spectrum domain.
• Development of a Support Vector Domain Description (SVDD) classifier.
• Optimization of SVDD parameters using the Particle Swarm Optimization (PSO) algorithm (PSO-SVDD).
• Introduction of negative examples (FOD samples) to prevent overfitting, creating a PSO-NSVDD classifier.

Main Results:

• The proposed PSO-SVDD and PSO-NSVDD classifiers demonstrated effective FOD detection capabilities.
• Experimental results confirmed good detection performance using measured radar data.
• A significant reduction in the false alarm rate was achieved with the proposed methods.

Conclusions:

• The PSO-optimized SVDD classifier, particularly with the inclusion of negative examples (PSO-NSVDD), offers a robust solution for FOD detection.
• The developed method enhances the reliability of millimeter-wave radar systems by improving classification accuracy and minimizing false alarms.
• This approach represents a significant advancement in automated FOD detection technology.