结合分类与迁移学习的薄云覆盖遥感图像地物信息恢复

利用多源多时相遥感图像，给出一种结合分类与迁移学习的薄云覆盖遥感图像地物信息恢复算法.首先利用多方向非抽样对偶树复小波变换对多源多时相遥感图像进行多分辨率分解，对分解后的薄云图像的高频系数利用贝叶斯方法进行地物初分类；再对每类地物的低频系数通过迁移最小方差支持向量回归模型进行域自适应学习，获取模型参数；最后利用所获的迁移回归模型，用无云参考图像的低频系数预测薄云覆盖图像的低频系数，去除薄云，恢复薄云覆盖图像的地物信息.实验结果表明，本文算法恢复的地物细节清楚，光谱失真较小.特别对地物季节性变化的薄云覆盖遥感图像，本文算法能有效恢复薄云覆盖区域的地物信息.

Ground Object Information Recovery for Thin Cloud Contaminated Remote Sensing Images by Combining Classification with Transfer Learning

By using multi-source and multi-temporal remote sensing images,a ground object information recovery algorithm for thin cloud contaminated remote sensing images is proposed by combining classification with transfer learning.Firstly,multi-resolution decomposition of multi-source and multi-temporal remote sensing images is performed by using multi-directional nonsubsampled dual-tree complex wavelet transform.The decomposed high frequency coefficients of the ground objects of the thin cloud images are primarily classified by using Bayesian method.Then the transfer least square support vector regression model is trained to obtain the model parameters by using the domain adaptive learning of the low frequency coefficients of each class of ground objects.Finally,the low frequency coefficients of the thin cloud-contaminated images are predicted by using those of the cloudless reference images.The thin clouds are removed and the ground object information of the thin cloud contaminated images is recovered.Experimental results show that the ground objects recovered by the proposed algorithm have clear spatial details and small spectral distortion.Especially for the thin cloud contaminated remote sensing images with seasonal variation of ground objects,the proposed algorithm can effectively recover the ground object information contaminated by thin clouds.