光学遥感邻近效应机理与模拟方法研究

项目名称： 光学遥感邻近效应机理与模拟方法研究

项目编号： No.61308098

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 无线电电子学、电信技术

项目作者： 马晓珊

作者单位： 中国科学院空间科学与应用研究中心

项目金额： 24万元

中文摘要： 随着光学遥感空间分辨率和应用遥感信息定量化程度要求的不断提高，开展光学遥感邻近效应机理与模拟方法研究已经成为定量遥感发展面临的重要前沿性科学难题，也是高空间分辨率光学遥感成像仿真高精度建模亟待解决的关键技术问题。邻近效应可以看作大气点扩散函数和地表辐射场的卷积，本项目针对传统蒙特卡罗模拟无法获得大气点扩散函数最优解的不足，采用BP神经网络，利用网络的后向传输监督学习机制，对特定传输条件下蒙特卡罗模拟光子传输得到的点扩散函数进行训练，获得稳定有代表性的大气点扩散函数，配合地表辐射场的计算结果，获得各种传输条件下的遥感器高度处的辐亮度值。在此基础上，分析邻近效应的影响范围，研究波长、像元尺度、地表反射率分布、关键大气参数及成像几何等对邻近效应的影响规律，为高分辨率光学遥感成像系统高精度建模和邻近效应校正算法研究奠定理论和技术基础，为光学遥感器的设计与优化提供科学依据。

中文关键词： 光学遥感；成像仿真；邻近效应；蒙特卡罗；BP神经网络

英文摘要： As the improvement of the spatial resolution of the optical remote sensing imaging and the development of the quantitative remote sensing, it has become a key problem to study the theory of remote sensing adjacency effect mechanism and factors, not only for the quantitative remote sensing, but also for the simulation of the optical remote sensing imaging system with high spatial resolution. Adjacency effect could be regarded as the convolution of the atmospheric point spread function and the surface-leaving radiance. Because the tranditional Monte Carlo simulation of the atmospheric point spread function can't assure the validity of the results, a BP neural network using a back-propagation supervised learning rule was introduced. By training the particular patterns of the point spread functions obtained by Monte Carlo simulation of the photons propagation under the specified propagation conditions, the neural network could estimate the atmospheric point spread function. Combined this results with the surface-leaving radiance, the at-sensor radiances under different propagation conditions were obtained. Based on the simulation results, the effective surface range was analyzed and the sensitivities of the adjacency effect to spectral regions,pixel size, nonuniform reflection surface, key atmospheric parameters and

英文关键词： Optical remote sensing；imaging simulation；adjacency effect；Monte Carlo；BP neural network