自适应全变分模型的高分辨率遥感影像房屋提取及后处理方法研究

The high-resolution remote sensing images are one of the most important data source to be adapted for the object detection. And more than half of the people in the world live in urban and suburban environments. Therefore, reliable and accurate detection of housing construction from high-resolution remote sensing images is a priority and a very active research field. The project around the characteristics of the shape of the housing construction in high-resolution remote sensing images showing complex diversity, size and variability, to carry out the extraction of housing construction, the description of the details of the housing structure and the precise positioning of the post-processing method. At first, the main research is focused on an adaptive multi-scale total variation segmentation model for housing construction objects initial extraction. In is step, the wealth of experience and knowledge of mankind are combination and which are simulated using the strictly mathematical model. And finally, cross-correction between the border and corner of housing is adapted for precise positioning based on the human experience and knowledge of the characteristics of their mutual relations. Then, a post-processing is taken for accurately extract the perfect geometry structure information of the housing construction object, including the smooth boundary, corner points and other characteristics. In is step, the wealth of experience and knowledge of mankind are combination and which are simulated using the strictly mathematical model. And finally, cross-correction between the border and corner of housing is adapted for precise positioning based on the human experience and knowledge of the characteristics of their mutual relations. The research results has a very important significance of academic value and practical significance in the precise positioning of the urban housing construction, the area of statistics, household management, urban development planning and management.

高分辨率遥感影像是地物目标探测一种非常重要的数据源。世界上超过一半的人生活在城市和近郊的环境中。因此,可靠、准确的房屋建筑探测是高分辨率遥感影像的一个非常重要且非常活跃的研究领域。 本项目围绕房屋建筑在高分辨率遥感影像中呈现出的形状复杂多样性、面积大小多变性等特点，开展房屋建筑的提取，房屋结构细节信息的描述及精确定位的后处理方法的研究。主要研究一种自适应多尺度全变分分割模型用于房屋建筑对象的初步提取；结合人类丰富的经验知识，采用严格的数学模型模拟房屋特征，进行房屋对象的后处理。以精确的提取房屋建筑对象规则的几何结构信息，包括光滑的边界，角点等特征。并依据房屋边界与角点的特征及其相互关系，进行交叉矫正并准确定位。 该研究成果在城镇房屋建筑的精确定位、面积统计、户籍管理，城镇发展规划与管理等方面有着极其重要的学术价值和现实意义。

高分辨率遥感影像是地物目标探测一种非常重要的数据源。世界上超过一半的人生活在城市和近郊的环境中。因此,可靠、准确的房屋建筑探测是高分辨率遥感影像的一个非常重要且非常活跃的研究领域。本项目围绕房屋建筑在高分辨率遥感影像中呈现出的形状复杂多样性、面积大小多变性等特点，开展房屋建筑的提取，房屋结构细节信息的描述及精确定位的后处理方法的研究。本课题首先提出了一种新的基于能量驱动的全变分分割方法，用于面向对象的建筑物区域的提取；由于变分算法中参数对结果影响很大，为了解决变分算法中最佳参数自动选择问题，使用了进化优化算法解决了该问题。为了快速自动建筑物区域的精确提取，还提出了一种基于数学形态学框架的建筑物区域从粗到细的提取方案。该研究成果在城镇房屋建筑的精确定位、面积统计、户籍管理，城镇发展规划与管理等方面有着极其重要的学术价值和应用意义。