A CNN-based Material Classification Approach Using Heatmap of a Dual-band Microwave Sensor
Nazli Kazemi, Mohammad Abdolrazzaghi, Petr Musı́lek, Elham Baladi · 2024
The performance of highly sensitive microwave res-onators deployed for material characterization is hindered by the lack of selectivity. This is especially noticeable when dealing with unknown chemicals. In this article, we propose a machine learning algorithm that enables selective sensing based on post-processing of the sensor response. A dual-band planar CSRR is designed at 1.45 GHz and 2.45 GHz, with sizes confined to 20% of the guided wavelength. It is used to discriminate between concentrations of 0-100% of ethanol/methanol in water, with increments of 20%. A convolutional neural network is used for multiclass classification with a high accuracy of 91.5%.