基于群不变稀疏表示的高分辨率SAR图像地物分类

The widely used local invariant features cannot adequately represent the complex region structure in high-resolution SAR images. Meanwhile the feature quantization for image representation is also a lossy process. To solve these problems, the model of group invariant feature extraction based on the framework of transformation groups and the factorized subgroups of network is constructed to achieve the robust representation of complex structures in images. The sparse coding theory is introduced into Fisher vector coding of the invariant feature, the method of sparse representation of the group invariant feature is built, and then sparse feature of group invariance is implemented, which can not only effectively represent the high dimensional feature produced by increasing number of network series, but also reduce the information loss of image representation. The obtained sparse feature is applied to the high resolution SAR image segmentation using weakly supervised semantic method based on the multi-image network model, based on which we research on the terrain classification in high resolution SAR images based on ensemble learning optimized by non-negative coding. This project will establish the foundation for the further development of theory and practice of the high-resolution SAR image processing field.

本项目针对目前广泛采用的局部不变特征不能充分表示高分辨率SAR图像中的区域复杂结构，以及图像表示过程中特征量化存在信息损失的问题，在群变换框架下利用网络子群分解特性构建群不变特征提取模型，实现对图像复杂结构的鲁棒表示。将稀疏编码理论引入不变特征的Fisher矢量编码问题，建立群不变特征的稀疏表示方法，得到群不变稀疏特征，达到有效表示随群变换网络级数增加而产生的高维特征的目的，同时降低图像表示的信息损失。所获得的群不变稀疏特征应用于多图像网络模型的弱监督语义高分辨率SAR图像分割，在此基础上研究非负编码优化的集成学习高分辨率SAR图像地物分类。本项目的研究将为高分辨率SAR图像处理领域的理论与实践进一步发展奠定一定的基础。

按照项目计划的安排，在稀疏Fisher矢量编码和深度神经网络基础上，本项目针对高分辨率SAR图像的分类问题展开研究，主要研究内容包括：稀疏Fisher矢量编码、优化学习Fisher矢量分类、非负约束稀疏编码Fisher矢量分类、深度网络与稀疏Fisher矢量图像分类、深度网络集成优化学习图像分类、深度网络高斯聚合编码图像分类、OpenCL并行加速图像分类等。同时，对于所研究的算法在图像分割和目标跟踪等领域进行了应用拓展。目前该研究已经基本按预定计划完成，培养博士生、硕士生13人，研究成果共计发表论文44篇，在《ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING》、《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》、《IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING》、《IEEE TRANSACTIONS ON FUZZY SYSTEMS》、《IEEE TRANSACTIONS ON CYBERNETICS》、《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》、《PATTERN RECOGNITION》等SCI检索的国际期刊发表论文38篇，其中SCI I区和II区论文30篇，国际会议论文4篇（SCI检索1篇CCF A类会议），申报国家发明专利19项（其中授权15项）。这些指标均已达到并超过研究计划中所列预期成果数量。