Atmospheric turbulence degradation image restoration based on multi-input multi-output U-Net network

作者：Rong Tao, Leihong Zhang, Yao Fang, Haima Yang, Dawei Zhang, Banglian Xu, Quan Sun · 发表于：Laser Physics · 年份：2025 · DOI：10.1088/1555-6611/ada750 · 被引用次数：2 · 研究领域：Advanced Image Processing Techniques、Image and Signal Denoising Methods、Optical Systems and Laser Technology

Abstract Aiming at the problem of limited ability to recover degraded images of atmospheric turbulence when they are severely damaged, a recovery method based on a multi-input multi-output U-Net network is proposed. The network can input information from multiple images at the same time, which improves the extraction and representation of image features. And it introduces multiple inverse convolutional head transposition attention during jump connection to better aggregate local and non-local pixel interactions and improve the recovery quality. Experiments demonstrate that the peak signal-to-noise ratio (PSNR) of the proposed network is improved by 4–5 db and the structural similarity is improved by about 6.4% compared with general recovery methods. Compared with the existing deep learning restoration methods, the PSNR of the improved network is improved by about 2–3 db, and the frequency structure similarity is improved by about 2.7%.