基于ScSPM-Reranking的高分辨率遥感影像的检索

摘要

为了从高分辨率遥感影像中获取详细的地表地物信息,为城市规划、环境监测以及灾情分析提供可靠的数据,进行了高分辨率遥感影像的检索研究,包括对图像的特征提取和图像之间相似度的描述.为了提高图像检索精度,运用了采用稀疏编码(Sc)的空间塔式匹配(ScSPM)技术和重排序(Reranking)技术,提出了基于ScSPM结合Reranking(ScSPM-Reranking)的遥感高分辨率影像的检索方法.该方法首先使用ScSPM提取空间场景的特征,然后结合这些特征使用cityblock距离进行初步检索,最后对初步检索的结果进行Reranking排序,获得高精度的检索结果.同其他检索方法进行了对比实验,实验结果证明,该方法具有较高的检索精度.%The study of remote sensing image retrieval, including image feature extraction and description of image similarity, was conducted to obtain the detailed information of the earth surface to provide reliable data for urban planning, environmental monitoring and disaster analysis.To improve the accuracy of image retrieval, the techniques of spatial pyramid matching (SPM) using sparse coding (Sc), called the ScSPM for short, as well as Reranking, were applied to the study, and a new method for retrieval of high resolution remote sensing images based on ScSPM and Reranking was proposed.The method uses ScSPM to extract the features of spatial scenes, then, uses the cityblock distance to perform the preliminary retrieval of database images, and lastly uses the Reranking method to boost the accuracy of image retrieval.The proposed method was compared with other retrieval methods by experiment and the method using ScSPM-Reranking achieved the higher retrieval accuracy.