基于神经网络的大气湍流退化图像的快速仿真  

Rapid Simulation of Atmospheric Turbulence Degradation Images Based on Neural Networks

作　　者：杨德荃 陶彦辉[2] 赵刚练 吴小龑 陈子阳 蒲继雄 YANG Dequan;TAO Yanhui;ZHAO Ganglian;WU Xiaoyan;CHEN Ziyang;PU Jixiong(Fujian Provincial Key Laboratory of Light Propagation and Transformation,School of Information Science and Engineering,Huaqiao University,Xiamen 361021,China;Institute of Fluid Physics,China Academy of Engineering Physics,Mianyang 621000,China;Unit 93184,Beijing 100071,China)

机构地区：[1]华侨大学信息科学与工程学院福建省光传输与变换重点实验室,厦门361021 [2]中国工程物理研究院流体物理研究所,绵阳621000 [3]93184部队,北京100071

出　　处：《航天返回与遥感》2023年第6期57-67,共11页Spacecraft Recovery & Remote Sensing

基　　金：中国工程物理研究院院长基金(YZJJZQ2022001);国家自然科学基金(62375092)。

摘　　要：大气湍流的随机扰动会对大气中传输的光场性质产生影响,导致远距离的目标经过大气湍流传输后出现图像退化和畸变。研究大气湍流的建模,对图像退化进行仿真模拟是矫正湍流引起的图像畸变、改善成像质量的关键,为此文章提出了一种快速实现仿真大气湍流所引起的图像畸变的方法,通过使用Zernike多项式表示大气湍流的相位,利用Zernike系数之间的相关性求解Zernike系数,搭建神经网络,实现了快速求解湍流的点扩散函数;依据图像退化的过程将清晰无失真图像模拟成受大气湍流影响的退化图像;同时,根据不同参数生成不同效果的湍流图像,分析大气湍流对成像的影响,并对斜程大气湍流传输对成像的影响进行了分析和讨论。此外,文章还将模拟图像与实际湍流图像进行了定量对比,两者的结构相似性(SSIM)和均方误差(MSE)指标数值仅差0.32%和0.92%,表明模拟图像与实际湍流图像具有较好的一致性,验证了文章所提方法在可见光波段应用的可行性。The natural fluctuations of the atmosphere,known as atmospheric turbulence,affects the properties of light waves,leading to image distortion and degradation of remote targets during propagation.Research on modeling atmospheric turbulence and simulating image degradation is crucial for correcting distorted images caused by turbulence and improving imaging quality.This study proposes a simulation method for rapidly implementing the image distortion caused by atmospheric turbulence.The phase of atmospheric turbulence is represented by Zernike polynomials,and a neural network is built by correlating Zernike coefficients to rapidly solve the point spread function of turbulence.Using this approach and employing a clear image as the input of the network,a degraded image affected by atmospheric turbulence can be simulated.We investigate the impact of key parameters such as the refractive index structure parameter,object distance and detector size on image degradation.Specifically,the image degradation is analyzed for an object as it propagates through slant-path atmospheric turbulence.To validate the proposed method,we quantitatively compare the simulated images to real turbulence ones,with the differences of only 0.32%and 0.92%in SSIM and MSE,validating the feasibility of the approach in the visible light band.

关 键 词：大气湍流 成像退化 湍流效应 斜程传输 

分 类 号：O439[机械工程—光学工程]