Millimeter-Wave Radar Sensing Using Deep Transfer Learning

Abstract

Millimeter-wave radar offers several advantages over traditional microwave radar, including higher resolution, greater accuracy, increased sensitivity to small targets, and less atmospheric attenuation. By utilizing deep learning techniques, the raw data can be processed to extract features and classify targets with high accuracy, making it ideal for sensitive applications. This work develops multiple new applications utilizing millimeter-wave FMCW radar. The first application monitors container vibrations to detect drilling vibrations autonomously. The system demonstrates the ability to detect micron-scale intrusive drilling at highway speed for the first time. The second application proposes the use of combined heart sound and gait signals for the first time as biometrics for human identification. Using the image augmentation technique and the joint probability mass function method, the two biometrics are combined to report a 98% identification accuracy. The third application demonstrate the improvement of the radar-based human activity recognition using the combination of four data domains: time-frequency, time-range, range-Doppler and, for the first time, time-angle domain. Six different activities are observed from nine subjects to achieve a combined classification accuracy of 100%. Lastly, the fourth application presents a human tracking system where three classifiers are utilized to identify the subject and their behavior. The system tracks the subject and detect the type of their motion. Based on the detected type of motion, the three classifiers are utilized for identification and activity recognition.