基于拉格朗日分解算法的SAR图像混合像元分解

Decomposition of SAR image mixed pixels based on lagrangian constrained neural network

为解决与光学遥感图像不同的合成孔径雷达(SAR)图像中存在大量混合像元的问题,本文提出了一种基于拉格朗日分解算法的SAR图像混合像元分解的方法,结合相关内容中具体定理的证明,文中给出拉格朗日分解算法用于SAR图像混合像元分解的系统的求解方法.用人工模拟SAR图像和ENVISAT SAR图像进行实验,结果表明拉格朗日分解算法的混合像元分解结果明显优于非约束类神经网络(文中实验以BP神经网络为例)的分解结果.

For resolving the problem of mixed pixels that the Synthetic Aperture Radar （SAR） image has which is different from optical remote sensing image, we apply the Lagrangian constrained neural network to decomposition of SAR image mixed pixels. Combining the demonstration of specific theorem in relevant content, we propose a systemic solving method which uses Lagrange constrained neural network decompose the mixed pixels of the SAR image. We make experiments on artificial simulated SAR images and ENVISAT SAR images. Experimental results show that the Lagrangian constrained neural network can get significantly more precise results than other neural network which does not contain restrictive conditions, （such as the BP neural network）.