Landmine detection and classification with complex-valued hybrid neural network using scattering parameters dataset.

Area of Science:

• Geophysics
• Artificial Intelligence
• Sensor Technology

Background:

• Landmine detection is crucial for safety and requires advanced sensor interpretation.
• Existing methods using real-valued neural networks with ground penetrating radar (GPR) often ignore vital phase information.
• Developing phase-sensitive detection techniques can significantly improve landmine identification accuracy.

Purpose of the Study:

• To introduce and evaluate complex-valued neural networks for landmine detection using GPR data.
• To leverage phase information for enhanced detection and classification of landmines.
• To propose a hybrid neural network architecture for comprehensive landmine analysis.

Main Methods:

• Utilized complex-valued neural networks to process GPR scattering parameters, incorporating phase information.
• Developed a two-layer hybrid neural network combining supervised and unsupervised learning.
• Tested the proposed network on a benchmark dataset for landmine detection and classification.

Main Results:

• Complex-valued neural networks demonstrated capability for phase-sensitive landmine detection.
• The hybrid network effectively detected and classified different types of landmines.
• Performance was validated using a standard benchmark dataset.

Conclusions:

• Complex-valued neural networks offer a superior approach to landmine detection compared to real-valued methods.
• Phase-sensitive analysis is critical for improving the accuracy of GPR-based landmine detection.
• The proposed hybrid network provides a robust framework for automated landmine identification.