High-performance multispectral ghost imaging based on the sine–cosine optimized patterns

Abstract

In recent years, the recovery of multispectral target scene has garnered increasing attentions from researchers, leading to the development of a series of ghost imaging schemes. However, the existing schemes still possess limitations such as requiring a large number of measurements and subpar performance. Therefore, here, we propose a deep-learning driven multispectral ghost imaging (MGI) scheme based on the sine–cosine optimized patterns (SCOP) for high-efficiency MGI. This scheme adopts a modified pattern selection strategy and relies on the powerful feature-extraction and representation-learning capabilities of multi-scale colour mapping (MSCM) network, which promise high-efficiency MGI for the multispectral target scenes. Experimental results show that the proposed MGI scheme can reconstruct complex multispectral target scenes with high quality at an ultra-low sampling rate (SR) of 2 %. In addition, the proposed scheme has excellent anti-noise performance and performs well in low signal-to-noise ratio (SNR) of 10 dB conditions. Overall, it provides a reliable solution for achieving fast high-quality MGI.