FMCW Radar-Based Material Characterization Using Convolutional Neural Network and K-Means Clustering

This paper proposes a machine learning model and a calibrated frequency-modulated continuous-wave (FMCW) radar sensor to characterize dielectric slabs. First, a calibration concept derived from vector network analyzer (VNA) measurements is used to calibrate the FMCW radar’s raw IF signal and to measure the reflection coefficient of a material at a much lower cost than the VNA. Second, the measured reflection coefficient is fitted to a complex-valued convolutional neural network (CNN) to determine the dielectric constant, loss tangent and thickness of the material. K-means clustering is proposed to reduce the complexity of the CNN by significantly reducing the number of classes. The results show that the proposed model enables the extraction of the material parameters with high accuracy.