项目名称: 基于语义模型的高分辨率卫星遥感图像人造目标检测方法研究
项目编号: No.61501460
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 无线电电子学、电信技术
项目作者: 李宇
作者单位: 中国科学院遥感与数字地球研究所
项目金额: 16万元
中文摘要: 目标检测一直是遥感图像应用领域的研究热点,随着空间分辨率的提高,遥感图像中目标细节日益精致,可辨识目标的种类和数量也大大增加。然而,分辨率的提高不仅丰富了有用信息,同时也凸显了目标的类内差异,还增加了背景干扰等信息,使得传统的遥感图像目标检测难以满足高分辨率卫星遥感图像信息提取的需要。本项目针对高分辨率卫星遥感图像的特点,在深入分析目标特征的基础上,采用层次化的语义表达形式,综合利用高分辨率遥感图像中的主题语义、形状特征、上下文关系等多类语义信息,搭建基于语义模型的目标检测框架;针对人造目标检测的具体应用,研究模型参数的优化,建立相应的模型学习推导理论;研究融合多类语义信息的模型构建及学习方法,最终实现复杂场景的高分辨率卫星遥感图像中目标的精确检测。本项研究将有效推动高分辨率卫星遥感图像在城市测绘、情报获取等方面的应用,并将有助于提升我国遥感的应用水平。
中文关键词: 高分辨率;卫星遥感图像;目标检测;语义模型;图像分割
英文摘要: Object detection has been a hot topic in the fields of remote sensing image application. With the progresses in spatial resolution of satellite sensors, the satellite remote sensing images contain more details of man-made objects. The number and species of identifiable objects increase greatly as well. However, the increasing amount of information not only highlights the details of interest objects, but also enhances all kinds of interference, so the existing object detection methods are not used to meet the needs of high resolution remote sensing images. Based on the depth analysis of the object features and hierarchical semantic expression, semantic model is constructed with the comprehensive utilization of the joint semantic topic, shape and context. Considering the practical detection tasks of man-made object, the optimal parameters are used to establish the corresponding model. The proposed method that fused multiclass information in remote sensing images is capable of dealing with man-made objects accurate detection problems within complex background. The research can promote the application level of high-resolution satellite remote sensing images in the area of urban mapping and information acquisition.
英文关键词: High spatial resolution;Satellite remote sensing images;Object detection;Semantic model;Image segmentation