Threat Assessment of Buried Objects Using Single-Frequency Microwave Measurements

Area of Science:

• Applied Physics
• Machine Learning
• Geophysics

Background:

• Buried object detection is crucial for security and unexploded ordnance (UXO) clearance.
• Existing methods often require complex systems or extensive training data.
• Microwave-based sensing offers a non-invasive approach to subsurface analysis.

Purpose of the Study:

• To develop a lightweight neural network model for buried object identification.
• To integrate this model with a microwave detection system using real-world measurements.
• To evaluate the model's performance against established deep learning architectures.

Main Methods:

• Utilized 16x16 scattering parameter (S-parameter) measurements from real-world data.
• Transformed S-parameter data into a 256-dimensional feature vector.
• Developed and trained a lightweight neural network architecture on the feature vectors.

Main Results:

• The model achieved 99.83% accuracy, 0.989 F1 score, and 0.979 recall in distinguishing hazardous objects.
• Outperformed baseline Convolutional Neural Network (CNN), Deep Residual Network (DRN), and EfficientNet models.
• Demonstrated robustness and practical relevance due to training on real-world measurements.

Conclusions:

• The proposed lightweight neural network and microwave detection system is highly effective for buried object identification.
• The approach offers a computationally efficient and accurate solution for defense and security.
• Real-world data integration enhances the model's reliability for practical deployment.