Learning-Based THz Multi-Layer Imaging for High-Capacity Positioning


This paper demonstrates a learning-based THz multi-layer pixel identification for contactless three-dimensional (3-D) positioning and encoders. More specifically, we propose a one-dimensional convolution-based residual network to deal with practical issues including 1) depth variation, 2) shadow effect, and 3) content recognition at the back surface of each layer. Experimental validation on a three-layer sample with contents on all surfaces is also provided.