Material Classification using 60-GHz Radar and Deep Convolutional Neural Network

In this paper we present a 60-GHz radar material classification system based on a Deep Convolutional Neural Network (DCNN), which uses images from the radar sensor as input features. These so called range cross-range images are generated with a Capon beamforming algorithm. As novelty for radar imaging classification, the neural network uses dilated convolution layers, which takes the physical characteristics of the radar images into account. Furthermore, the image augmentation technique helps to increase the accuracy, and besides reduces the amount of needed training data. A classification accuracy of above 97 % is achieved while classifying 1875 radar images from ten different materials. We also show a strong robustness against sensor noise and device nonlinearities, which is an advantage of the used radar feature imaging technique.