An Efficient Optimization Method for Microwave Sensors Based on Pixelation Design and a Convolutional Neural Network

This study leverages the pixelation design and a convolutional neural network to optimize the structure of a microwave sensor. The proposed approach enables efficient exploration of complex geometric configurations without manual intervention, overcoming limitations of traditional evolutionary algorithms. Applied to a complementary split-ring resonator microfluidic sensor, the method achieves a sensitivity of 0.69%, which is a 213% improvement over the baseline design. Experimental results demonstrate the framework's capability to generate unconventional but high-performance sensor geometries while significantly reducing computational costs compared to conventional optimization techniques. The proposed approach offers a generalizable paradigm for microwave sensor optimization that could be extended to other electromagnetic device designs.