结合分类与迁移学习的薄云覆盖遥感图像地物信息恢复

发布时间：2018-07-09 12:28

本文选题：遥感图像 + 信息恢复　；参考：《电子学报》2017年12期

【摘要】：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.
[Abstract]:Based on multi-source and multi-temporal remote sensing images, an algorithm for restoration of ground object information in thin cloud overlay remote sensing images is presented, which combines classification and migration learning. Firstly, multi-direction non-sampling dual tree complex wavelet transform is used to decompose multi-source and multi-temporal remote sensing images, and Bayesian method is used to classify ground objects for the high frequency coefficients of the decomposed thin cloud images. Then, the low frequency coefficients of each kind of objects are learned by the least variance support vector regression model, and the parameters of the model are obtained. Finally, the migration regression model is used. The low frequency coefficients of the cloudless reference image are used to predict the low frequency coefficients of the thin cloud overlay image, to remove the thin cloud, and to restore the ground object information of the thin cloud overlay image. The experimental results show that the restored features of the algorithm are clear and the spectral distortion is small. Especially for the remote sensing image of thin cloud covering with seasonal variation of ground objects, this algorithm can effectively restore the ground object information of thin cloud coverage area.
【作者单位】：安徽大学计算智能与信号处理教育部重点实验室;安徽大学电子信息工程学院;偏振光成像探测技术安徽省重点实验室;
【基金】：国家自然科学基金(No.61672032,No.61401001) 安徽省自然科学基金(No.1408085MF121) 偏振光成像探测技术安徽省重点实验室开放课题(No.2016-KFKT-003)
【分类号】：TP751

【相似文献】