Convolutional Neural Network Approach to THz Reflection Alignment

A major obstacle to non-contact measurement of in-vivo skin samples is the alignment sensitivity of reflected THz pulses. Movement of human skin on the relevant length scales of 10 mu m is faster than the movement capabilities of current state of the art medical robots: this means accurate non-contact alignment when imaging human skin is currently impossible with mechanical means. Presented here is a promising neutral network approach to correct the pulse from a misaligned terahertz system. This low-cost alternative to mechanical alignment correction will contribute to the development of in-vivo biomedical applications of THz imaging.