“数据-知识”驱动的大区域高分辨率遥感影像多尺度分割并行计算方法

项目名称： “数据-知识”驱动的大区域高分辨率遥感影像多尺度分割并行计算方法

项目编号： No.41501453

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 天文学、地球科学

项目作者： 周亚男

作者单位： 中国科学院地理科学与资源研究所

项目金额： 20万元

中文摘要： 影像多尺度分割是面向对象的遥感信息提取与目标识别的前提，其分割效果和计算效率直接关系着后续遥感分析的精度与应用水平；但当前影像多尺度分割算法的结果精度有限、且计算效率低下，难以满足需求日益强烈的大区域精细化遥感应用。针对遥感影像多尺度分割效果与效率问题，本项目将以影像多尺度空间表达为切入点，探讨先验知识的在多尺度分割中的融入、对象复杂度的计算等问题，发展和完善“数据—知识”双向驱动的影像多尺度分割模型，提高遥感影像多尺度分割的精度；以影像分割的多尺度特性为突破口，研究遥感影像数据的树型分块、多尺度分割的自适应停止判定等具体问题，建立“自底向上”和“自顶向下”相结合的高分辨率遥感影像多尺度分割的并行计算框架，突破数据分块和计算并行的技术限制，着力提升影像多尺度分割的计算效率；进而以省级大区域精细地类图斑的快速生成为应用出口，提高高分辨率遥感应用的有效性与规模化。

中文关键词： 高分辨率影像；专题知识；多尺度分割；并行计算；大区域遥感

英文摘要： Multi-scale image segmentation is the foundation and precondition for object-oriented remote sensing analysis of geo-information extraction and target recognition, its results and efficiency are directly relevant to accuracy and application of the following geo-spatial analysis. While the current methods of image segmentation with limited precision and low efficiency, are difficult to meet the increasing demand of large-region and precise applications. Aim at solving the problem of limited precision and low efficiency in multi-scale segmentation, our project will take the multi-scale representation of remote sensing image as a point cut, and discuss how to incorporate prior knowledge into segmentation and how to describe the complexity of a segmented object, and propose a data & knowledge driven model for multi-scale segmentation, to improve the accuracy of image segmentation. We will take multi-scale characteristic of image as a breakthrough point, and study the data partition and the finishing point in parallel computation of segmentation, and build a bottom-up & top-down incorporated framework for parallel computation of segmentation, to improve the calculation efficiency of image segmentation. In practice, the above algorithms are implemented for rapid and fine generation of segmented object in large-region images, enhancing validity and effectiveness of high-resolution remote sensing application.

英文关键词： high-resolution image;thematic knowledge;multi-scale segmentation;parallel computing;large-region remote sensing