Single-Pixel Hyperspectral Imaging via an Untrained Convolutional Neural Network

• DOI: 10.3390/photonics10020224
• Publication Date: 2023-02-20T07:29:08Z
• Authors: Chen-Hui Wang; Hong-Ze Li; Shu-Hang Bie; Rui-Bing Lv; Xi-Hao Chen
• Keywords: 0301 basic medicine; 03 medical and health sciences; hyperspectral imaging; single-pixel imaging; untrained neural network; deep learning; deep image prior; Applied optics. Photonics; TA1501-1820

Single-pixel hyperspectral imaging (HSI) has received a lot of attention in recent years due to its advantages high sensitivity, wide spectral ranges, low cost, and small sizes. In this article, we perform single-pixel HSI experiment based on an untrained convolutional neural network (CNN) at ultralow sampling rate, where the high-quality retrieved images target objects can be achieved by every visible wavelength light source from 432 nm 680 nm. Specifically, integrate physical model into randomly initialized CNN, which allows reconstructed relying solely interaction between process without pre-training network.