Highly precise optical positioning through deep learning-assisted photo-thermoelectric detection

Abstract

Optical positioning through a position-sensitive detector (PSD) is a central technique for diverse applications, including laser guidance, pilotless automobiles, aerospace, real-time tracking, and robotics. However, both commercially segmented PSD and the lateral photovoltaic effect PSD reported so far suffer from serious nonlinearity, leading to distortion of position. Herein, we propose an ultrahigh-precise optical positioning method through a deep learning-assisted thermoelectric model. Specifically, we design a PbSe thermoelectric photodetector with twelve electrodes endowing the output of the photo-thermoelectric voltage matrix, which is highly laser position-correlated. As a result, the original location is accurately reconstructed through a regression-based convolutional neural network algorithm, giving rise to a nonlinearity of 0.3% with precision as high as 98.7%. Finally, our thermoelectric deep optical positioner fully recovers the moving path of a target defined by alphabets, numbers, or random trajectory, indicating its great potential in the applications of optical writing, robot guidance, and optical navigation.