GWO-Based Joint Optimization of Millimeter-Wave System and Multilayer Perceptron for Archaeological Application

Area of Science:

• Archaeological science
• Radar imaging
• Machine learning

Background:

• Terahertz (THz) radar is a novel imaging modality for archaeological applications.
• Classifying pottery shards is vital for tracing the spread of agriculture from the Fertile Crescent to Europe.
• Optimizing sensor count in radar systems is key for cost-effectiveness and compactness.

Purpose of the Study:

• To jointly design a THz radar system and a classification neural network for pottery shard analysis.
• To minimize the number of sensors and associated costs while maintaining high classification accuracy.
• To reduce the false recognition rate in pottery shard classification.

Main Methods:

• Development of a novel Continuous Binary Ternary Grey Wolf Optimizer (CBTGWO).
• Joint optimization of radar system design and neural network classification.
• Sensor selection and placement strategies using evolutionary algorithms.

Main Results:

• In a 7-frequency scenario, CBTGWO reduced 37 sensors to 1 with a 0% false recognition rate.
• In a single-frequency scenario, CBTGWO selected 2 out of 217 sensors with a 2% false recognition rate.
• Acquisition time was reduced to 3.2 seconds, significantly outperforming existing GWO methods.

Conclusions:

• The CBTGWO effectively minimizes sensor requirements for THz radar-based archaeological classification.
• This approach offers a cost-effective and efficient solution for analyzing pottery shards.
• The method has strong potential for advancing our understanding of historical agricultural diffusion.