基于高分影像的城市空间结构场景识别方法研究

Traditional information extraction for high-resolution remote sensing images can’t handle various application requirements of urban with complex human activities. At the same time, the rapid developments of China’s urbanization make it indispensable to recognize both the morphology and macro-structure of urban region. Based on high-resolution remote sensing images, combined with a series of methods for image scene recognition, this project proposes to construct a block-oriented semantic classification model for urban scenes and then automatically establish fine spatial structures of urban scenes. Furthermore, it tries to have a research on algorithms of division of urban functional zones and form a perfect methodology system of scene recognition of urban spatial structures, so that rapid, fine division and comprehensive analysis on a variety of urban functional zone can be carried out. To begin with the research on sematic types and expression means of urban scenes, the project focus on the key issue about the feature extraction and construction of a visual dictionary for irregular image blocks so as to break through technological bottlenecks that scene classification algorithm of remote sensing images can only be applied to regular image blocks. Moreover, utilizing the fusion of the image scene classification method based on low-level features and the semantic classification method based on visual vocabulary, this project will develop brand new block-oriented scene classification algorithm for remote sensing images. The positive results this project achieves are expected to promote the construction of new urban spatial structures on a microscopic scale supported by high-resolution remote sensing images and greatly contribute to the application of evolution rules and development mechanisms of urban spatial structures.

在人类活动复杂的城市区域，传统高分遥感信息提取技术无法满足其应用需求，而我国城镇化的快速发展又急需全面了解城市空间的微观形态，并在此基础上掌握城市的宏观结构。本项目立足高空间分辨率影像，结合图像场景识别的最新方法，构建以城市街区为基本场景单元的遥感影像场景分类模型，实现城市场景精细空间结构的自动识别，并进一步研究城市功能区划分算法，形成完善的多层次城市空间结构场景识别方法体系，满足多种类型城市功能区大范围、快速、精细划分与综合分析。拟从研究城市场景语义类型及表达方式着手，重点解决不规则图斑特征提取及视觉字典构建的技术难题，突破以往遥感影像场景分类方法只能作用于规则影像块的技术瓶颈，联合发展基于图像低层特征的场景分类方法和基于图像视觉词汇的语义分类方法，建立街区式遥感影像场景分类算法。本项目将推动高分遥感支持下的微尺度新型城市空间结构研究的发展，促进城市空间结构演化规律、动力机制等的深入研究

本项目科学问题是城市复杂格局要求精细化表达、多尺度分析，与高分遥感信息维度缺乏、自动化智能化技术实现困难之间的矛盾。为此项目以米级高分影像为主体，从基于像素、对象与场景不同角度，发展一套面向城市土地资源的多源信息协同下最优分割、高维特征分类、要素与场景信息提取的方法和技术体系；构建城市基底（行政界线-建成区-街区）—城市廊道（道路网、水系）—城市斑块（人工、自然地表要素）的多层级架构和精细地表要素产品制作关键技术实现；在精细化表达基础上开展多尺度单元，针对不同城市从景观空间分布格局和城市空间功能等角度开展对比分析与演化模式应用研究。该项目为实现城市精细地表要素信息提取提供了新方法，拓展了新思路。并通过方法体系构建和应用实例展现了遥感大区域范围下多表现形态的典型场景提取与分析应用的重要性和实用性。项目主要研究内容包括五个方面：高分遥感影像多尺度分割关键尺度参数选择与等级斑块构建与应用、米级高分数据城市建成区轮廓提取与精细地表覆被制图、城市场景的分割、分类及典型场景识别提取、基于深度学习的大区域场景监测应用、街区单元的典型城市功能格局分析与要素演变综合应用。形成实现的大区域典型城市及人工居民地提取、标识和场景解译方法与技术体系，满足多种类型城市要素及功能区大范围、快速、精细划分，以及基于街区空间单元的空间格局综合分析。研究成果在智慧城市、高分影像智能信息提取、多源遥感动态监测与多源信息智能计算等方面都具有广阔的应用前景。