A Feasibility Study of Microwave Metal and Dielectric Classification Based on Machine Learning Techniques

This paper studies the classification of metal and dielectric based on microwave signals using machine-learning techniques. Numerical simulations are applied to generate synthetic data. Training samples with metallic and dielectric objects are simulated, respectively. The synthetic data are then applied to the training and testing of Supporting Vector Machine (SVM) and Convolutional Neural Network (CNN). The average accuracy of SVM and BP is 82% and 98 %, respectively. The results indicate the feasibility of microwave metallic and dielectric targets classification using learning-based techniques, which can provide more a priori information for the subsequent imaging algorithms.