中文摘要：

针对现有分布式影像边缘提取方法中因数据集划分不合理引起的影像特征丢失问题，提出一种弹性影像金字塔（resilient image pyramid, RIP）模型，采用多层冗余的数据剖分保证原始影像特征的完整性，通过关键元数据对金字塔进行存储，减少计算过程中的数据迁移量。在此基础上，设计并实现了一种适用于海量遥感影像的MapReduce计算方法，通过Hadoop进行分布式边缘提取。实验结果表明，采用此方法进行边缘提取的影像特征提取率得到了一定提升，是一种较为有效的保障提取结果完整性的计算方法。

英文摘要：

To resolve the problem that image features lost in distributed edge extraction caused by the impropriety dataset split algorithm, a subdivision model called resilient image pyramid （RIP） model is presented. With the multi-layer and redundant hi- erarchy of RIP, the integrity of original image features is assured. And the data migration is reduced, since RIP is stored in the form of critical metadata, which is very small. Based on RIP, a MapReduce method for edge extraction of massive remote sensing images is designed and implemented, which realizes the distributed edge extraction with Hadoop. Compared to traditional meth- od, the extraction rate of image feature using this method is improved; it is an efficient method to improve the extraction rate of distributed edge extraction.