网格梯度曲率模型零度Metropolis-Hastings遥感图像缺失重建

Zero degree Metropolis-Hastings with grid gradient-curvature model for remote sensing image reconstruction

针对矩形网格遥感图像可微随机域数据缺失重建过程中,存在重建效果不佳且计算效率不高的问题,提出一种网格化全局几何约束零度Metropolis-Hastings遥感图像缺失数据随机重建算法。首先,构建遥感图像的随机梯度-曲率重建模型,通过全局几何约束相互作用随机场模型,匹配整个网格样本的梯度和曲率,从而满足蒙特卡罗模拟应用条件;其次,采用蒙特卡罗算法改进版本零度Metropolis-Hastings算法,实现遥感图像缺失数据重建,该方式不承担对底层数据的概率分布参数描述,有助于降低用户参与度,提高计算效率,适用大型遥感图像数据集的无监督自动处理;最后,通过与其他分类或插值方法实验对比显示,所提算法在数据重建效果和计算效率上均要优于对比算法。

In order to solve the problem of poor performance for data reconstruction and low efficiency for algorithm computation in micro domains randomly missing data reconstruction process of remote sensing image, the zero degree Metropolis-Hastings with grid gradient-curvature model for remote sensing image reconstruction is proposed. Firstly, the stochastic gradient curvature reconstruc- tion model is constructed for remote sensing image, through the global geometric constraints interaction random field model, and the gradient and the curvature of the whole grid sample can be matched to meet Monte Carlo simulation application conditions. Sec- ondly, the zero metropolis Hastings algorithm is used, which is the improved version of Monte Carlo algorithm, to realize the remote sensing image missing data reconstruction, this way do not bear on the underlying data probability distribution parameters descrip- tion, which help to reduce the user participation and improve the computational efficiency, and is applicable to large remote sens- ing image. Finally, by comparing with other classification or interpolation method, the proposed algorithm is superior to the con- trast algorithm in data reconstruction and computation efficiency.