Fast optical‐writing recognition based on two‐dimensional photothermoelectric effect assisted with deep learning

Abstract

Abstract The photothermoelectric (PTE) effect enables a simple structure to construct a noncontact and self‐powered writing device. PTE detectors based on active materials and one‐dimensional (1D) position‐sensitive PTE effect have attracted considerable attentions. However, there is no research on applying the 1D position‐sensitive PTE effect to a two‐dimensional (2D) surface. Besides, flexible writing devices based on the PTE effect have not been reported. Here, we demonstrate a self‐powered flexible optical writing device based on the 2D position‐sensitive PTE effect of SnSe thin film. Laser positioning, tracking, and distinguishing of optically written characters have been realized by only measuring the bi‐directional output voltages during the writing process. A deep convolutional neural network has been introduced to recognize optically written numbers and characters with high recognition accuracy (92%) and fast processing speed (8.6 ms). The flexible optical‐writing device with a simple structure could rapidly and accurately recognize characters in a noncontact fashion using a small amount of training data.

Keywords

• deep neural network
• optical writing
• photothermoelectric effect
• tin selenide