Microwave Imaging of Human Body from Point Cloud with Capsule Network

3D microwave imaging of human targets has potential applications in military detection and intelligent interaction fields, we propose an auto-encoder-like capsule network to complete this task. We make two different 3D microwave imaging datasets MI-1 and MI-2 for verifying this work, the human action in MI-1 changes greatly, and the human action in MI-2 changes slightly. Experimental results show that our capsule network achieves nice results on both MI-1 and MI-2 datasets.