【作者】 王涛；

【导师】 洪津； 易维宁；

【作者基本信息】 中国科学技术大学 ， 光学， 2021， 博士

【摘要】 大气分子和气溶胶粒子对太阳辐射的吸收和散射会导致可见近红外波段卫星影像变得模糊、低对比度,并且还会导致卫星影像中各像元表观反射率严重偏离其地表真实反射率。研究大气校正技术对提升遥感图像质量、准确获取地表辐射收支、生物理化参量及水循环等地球资源信息至关重要。文章针对亚米级空间分辨率光学卫星遥感图像特点,分析了大气吸收、大气程辐射、邻近效应等对其的影响,提出了自适应大气校正算法。可见近红外波段卫星影像大气校正主要包括大气吸收校正、大气程辐射校正和邻近效应校正。大气吸收校正的本质是计算太阳辐射传输路径(太阳-地球表面-星载光电传感器)上的大气粒子对太阳辐射的总吸收透过率。仿真分析结果表明,提升大气吸收校正精度的关键在于获得与卫星影像时空同步的水汽和臭氧含量。大气程辐射校正的本质是计算大气程辐射值(或大气本征反射率)。目前主要利用黑暗像元法或辐射传输方程法来获得卫星影像中各像元对应的大气程辐射值。仿真分析结果表明,提升大气程辐射校正精度的关键在于获得与卫星影像时空同步的气溶胶类型及气溶胶含量。邻近效应校正的本质是计算各目标像元对应的平均背景反射率,所以邻近效应校正的关键在于确定邻近效应水平范围以及该范围内各背景像元对平均背景反射率的贡献权重值。但是目前缺乏针对邻近效应水平范围以及该范围内各背景像元对平均背景反射率的贡献权重值的相关影响因素的系统全面地分析。因此本文通过理论分析结合仿真数据,系统性地分析了邻近效应校正的相关影响因素。分析结果表明观测波长越小、空间分辨率越高、观测天顶角越小、地球表面目标物海拔高度越高、大气分子散射层高度越低、气溶胶散射层高度越低、大气分子光学厚度越大、气溶胶光学厚度越大,有效邻近效应水平范围越小;大气分子光学厚度越大、气溶胶光学厚度越大、背景像元与目标像元空间距离越小、背景像元反射率与目标像元反射率差值越大,该背景像元对平均背景反射率的贡献权重值越大。亚米级空间分辨率光学卫星影像中相邻地物反射率差异较大是导致邻近效应严重的一个主要因素,目前常用的大气校正算法均没有考虑背景像元反射率与目标像元反射率差值对计算该背景像元对平均背景反射率的贡献权重值的影响,故传统大气校正算法无法对亚米级空间分辨率卫星影像进行有效邻近效应校正。本文首次考虑了背景像元反射率与目标像元反射率差值对邻近效应贡献权重值的影响。并根据仿真分析结果,利用卫星入瞳处背景像元辐亮度和卫星入瞳处目标像元辐亮度的比值来定量描述背景像元反射率与目标像元反射率差值对计算背景各像元对邻近效应贡献权重值的相对大小。并将该比值和6S模型中的大气校正算法(6S-AC)中的平均背景反射率表达式结合得到等效平均背景反射率表达式,然后利用等效平均背景反射率替代6S-AC中的平均背景反射率,开发了自适应大气校正算法(Adaptive-AC)以用于亚米级空间分辨率光学卫星影像的大气校正。与6S-AC中的平均背景反射率相比,本文提出的等效平均背景反射率同时考虑了背景像元反射率与目标像元反射率的差值、气溶胶光学厚度、背景像元与目标像元之间的空间距离以及大气分子光学厚度对计算各背景像元对邻近效应的贡献权重值的影响。自适应大气校正算法的自适应特点体现在它可根据大气分子光学厚度、气溶胶光学厚度、背景像元与目标像元之间的空间距离和反射率差值来调整各背景像元对邻近效应的贡献权重值。利用Adaptive-AC对WorldView-3全色波段卫星影像(空间分辨率为0.31 m)和GF-2全色波段卫星影像(空间分辨率为0.81 m)进行大气校正,并与6S-AC和MODTRAN模型中的大气校正算法(MODTRAN-AC)的大气校正结果作对比,结果表明Adaptive-AC的校正结果优于6S-AC和MODTRAN-AC的校正结果。以2020年3月20日嵩山定标场的GF2全色波段卫星影像为例,高反射率靶标区域的地面同步实测反射率值为0.4756,低反射率靶标区域的地面同步实测反射率值为0.0681。当取邻近效应范围值为1920米时,基于Adaptive-AC校正后的卫星影像(简记为“Adaptive-AC地表真实反射率图”)中各靶标区域的平均反射率非常接近地面同步实测的平均反射率。Adaptive-AC地表真实反射率图中高反射率靶标区域的平均反射率为0.4406,低反射率靶标区域的平均反射率为0.0805;基于6S-AC校正后的卫星影像(简记为“6S-AC地表真实反射率图”)中各靶标区域的平均反射率与地面同步实测的平均反射率差别较大。6S-AC地表真实反射率图中高反射率靶标区域的平均反射率为0.3913,低反射率靶标区域的平均反射率为0.1004;基于MODTRAN-AC校正后的卫星影像(简记为“MODTRAN-AC地表真实反射率图”)中各靶标区域的平均反射率均高于地面同步实测值。MODTRAN-AC地表真实反射率图中高反射率靶标区域的平均反射率为0.4880,低反射率靶标区域的平均反射率为0.1211。Adaptive-AC的校正结果与常规图像处理结果作对比,结果表明相对于常规图像处理方法在提升卫星影像质量时会带来噪声和过度增强的问题,Adaptive-AC在提升卫星图像质量时不会带来噪声和过度增强问题。以2018年4月1日北京万泉河桥附近区域的WV-3全色波段卫星影像为例,从基于Adaptive-AC校正后的卫星影像可直接目视解译出大型车辆的车顶天窗,但从常规图像处理后的卫星影像中无法目视解译出这些信息。总的来说,与6S-AC中的平均背景反射率相比,本文提出的等效平均背景反射率可以更好地描述邻近效应的基本特征。并且基于等效平均背景反射率开发的Adaptive-AC可以很好地移除亚米级空间分辨率光学卫星对地球表面目标成像过程中,大气和目标物周围自然环境对目标物成像的影响,恢复卫星成像过程的真实性,提高亚米级空间分辨率卫星影像的图像质量和定量遥感精度。相对于MODTRAN-AC和6S-AC,Adaptive-AC更适合用于亚米级空间分辨率光学卫星影像的大气校正。更多还原

【Abstract】 The absorption and scattering of solar radiation by atmospheric molecules and aerosol particles will cause the visible and near-infrared(VNIR)satellite images to become blurred and low-contrast,and also cause the apparent reflectance of each pixel in the satellite image to deviate from its true reflectance on the earth surface.Research on atmospheric correction technology is essential to improve the quality of remote sensing images and accurately obtain information on earth resources such as surface radiation budgets,biophysical and chemical parameters,and water cycles.Based on the characteristics of sub-meter spatial resolution optical satellite images,this dissertation analyzes the influence of atmospheric absorption,atmospheric path radiation and adjacency effect on atmospheric correction.Based on the above analysis results,an adaptive atmospheric correction algorithm is proposed.For the VNIR satellite images,atmospheric correction mainly includes atmospheric absorption correction,atmospheric path radiance correction and adjacency effect correction.The essence of atmospheric absorption correction is to calculate the total absorption transmittance of the atmospheric particles to solar radiation on the solar radiation transmission path.The simulation analysis results show that the key to improving the accuracy of atmospheric absorption correction is to obtain the atmospheric water vapor and ozone content which is time-space synchronized with satellite image.The atmospheric path radiation correction is essentially to calculate the atmospheric path radiation value.At present,the dark pixel method or the radiation transfer equation method is mainly used to obtain the atmospheric path radiation value corresponding to each pixel in the satellite image.The simulation analysis results show that the key to improving the accuracy of atmospheric path radiation correction is to obtain the atmospheric aerosol types and aerosol optical depth which is time-space synchronized with satellite image.The adjacency effect correction aims at solve the average background reflectance(ABR)of each pixel in the satellite image.Therefore,the key to the adjacency effect correction is to determine the value of the adjacency effect range and the contribution factor(or weight value)of the background pixel in this range to the ABR.However,there is currently a lack of systematic and comprehensive analysis of the related influence factors of the adjacency effect range and the contribution weight value of each background pixel to the ABR.Therefore,this dissertation systematically analyzes the related influence factors of adjacency effect correction through theoretical analysis and simulation data.The analysis results show that the smaller the observation wavelength,the higher the spatial resolution,the smaller the observation zenith angle,the higher the altitude of the target on the surface of the earth,the lower the height of the atmospheric molecular scattering layer,the lower the height of the aerosol scattering layer,the greater the optical thickness of atmospheric molecules and the greater the aerosol optical thickness,the smaller the effective adjacency effect range value.The greater the optical thickness of atmospheric molecules,the greater the aerosol optical thickness,the smaller the spatial distance between the background pixel and the target pixel and the greater the reflectance difference between the background pixel reflectance and the target pixel reflectance,the greater the contribution factor(or weight value)of the background pixel to the ABR.The large reflectance difference of adjacent ground objects in sub-meter spatial resolution optical satellite images is a major factor leading to serious adjacency effects.However,the traditional atmospheric correction algorithms do not consider the effect of RDCS on the ABR.These algorithms become inappropriate to correct the adjacency effect in SM satellite images.In this dissertation,for the first time,the influence of the reflectance difference between the background pixel and target pixel on the contribution weight values of the adjacency effect is considered.And the simulation analysis results show that the ratio of the radiance of the background pixel at the entrance pupil of the satellite to the radiance of the target pixel at the entrance pupil of the satellite(Lback/Ltarget)can quantitatively represent the relative magnitude of the contribution weight value of RDCS to the ABR.Based on this situation,we combine the value of Lback/Ltarget with the ABR expression in the 6S-AC to obtain the equivalent average background reflectance(EABR)expression,and replace ABR in the formula of the 6S-AC with EABR.The EABR depends on the the atmospheric molecules optical thickness,aerosol optical thickness,the spatial distance from the central pixel to its surrounding pixels(SDCS),and RDCS.The advantage of EABR over ABR is that the former considers contributions from the the atmospheric molecules optical thickness,aerosol optical thickness,SDCS,and RDCS to the adjacency effect.EABR can better represent the adjacency effect in the SM satellite image.Here,we develop an adaptive atmospheric correction algorithm(adaptive-AC)based on EABR to perform atmospheric correction on SM satellite images.Adaptive-AC can adjust the contribution factor(or weight value)of the background pixel to the adjacency effect according to the atmospheric molecules optical thickness,aerosol optical thickness,SDCS,and RDCS.The application of adaptive-AC to WorldView-3 panchromatic band satellite imagery(spatial resolution of 0.31 m)and GF-2 panchromatic band satellite imagery(spatial resolution of 0.81 m)eveals that adaptive-AC can significantly improve image quality,with the reflectance of pixels in the corrected image processed by adaptive-AC being close to the actual reflectance.The results show that the correction results of Adaptive-AC are better than those of 6S-AC and MODTRAN-AC.Taking the GF2 panchromatic satellite image of the Songshan calibration field on March 20,2020 as an example,the ground synchronous measured reflectance value of the high reflectance target area is 0.4756,and the ground synchronous measured reflectance value of the low reflectance target area is 0.0681.When the adjacency effect range value is 1920 meters,the average reflectance of each target area in the satellite image corrected based on Adaptive-AC(abbreviated as "Adaptive-AC true surface reflectance image")is very close to the average reflectance measured on the ground.The average reflectance of the high reflectance target area in the Adaptive-AC true surface reflectance image is 0.4406,and the average reflectance of the low reflectance target area is 0.0805;The average reflectance of each target area in the satellite image corrected based on the 6S-AC(abbreviated as "6S-AC true surface reflectance image")is quite different from the average reflectance measured on the ground.The average reflectance of the high reflectance target area in the 6S-AC true surface reflectance image is 0.3913,and the average reflectance of the low reflectance target area is 0.1004;The average reflectance of each target area in the satellite image corrected based on the MODTRAN-AC(abbreviated as " MODTRAN-AC true surface reflectance image”)is higher than the reflectance measured on the ground.The average reflectance of the high reflectance target area in the MODTRAN-AC true surface reflectance image is 0.4880,and the average reflectance of the low reflectance target area is 0.1211.Comparing the correction results of Adaptive-AC with the results of conventional image processing indicates that conventional image processing methods will bring noise and over-enhancement problems when improving the quality of satellite images,while Adaptive-AC will not bring noises and over-enhancement problems when improving the quality of satellite images.Taking the WV-3 panchromatic satellite image of the area near the Wanquan River Bridge in Beijing on April 1,2018 as an example,the roof sunroof of a large vehicle can be directly visually interpreted from the satellite image corrected based on Adaptive-AC.However,the information cannot be visually interpreted from the satellite image after conventional image processing.In general,that compared with ABR,EABR can better describe the essential characteristics of the adjacency effect,and the proposed corresponding adaptive-AC can well remove the effect of the atmosphere and surrounding environment from an SM satellite image,to restore the authenticity of the satellite imaging process,and improve the image quality and quantitative remote sensing accuracy of SM satellite image.Adaptive-AC is more suitable for SM satellite image preprocessing than 6S-AC and MODTRAN-AC approaches.更多还原