项目名称: 形状先验和数据驱动的高分辨遥感影像目标提取
项目编号: No.41471280
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 天文学、地球科学
项目作者: 汪西莉
作者单位: 陕西师范大学
项目金额: 81万元
中文摘要: 高空间分辨率遥感影像提供了更为丰富、精细的地面信息,但利用源于图像数据的光谱、纹理、边缘等局部特征,在有遮挡、阴影、噪声影响、背景杂乱、多目标等复杂场景下,依然难以自动提取出完整的目标。鉴于遥感目标,特别是人工地物具有特定的事先已知的形状,项目提出基于遥感、统计、图像配准、基于图的图像分割、分类等领域的理论和方法,针对城市高分辨遥感影像中建筑物、道路、绿地等目标,以图像数据为基础,以与目标相似的形状作为先验知识,研究形状的统计表示模型和具有单个形状模板以及多个形状模板的图像目标提取方法,利用形状先验和图像数据解决复杂情况下的目标提取问题。项目针对矩形、具有平行边的长条状、交叉路口等形状,系统地研究形状模型和全局形状特征作为先验知识,与图像特征共同作用提升高分辨遥感影像目标提取性能的方法,对深化先验知识的研究、基于形状先验和图像数据的遥感目标提取方法的研究和应用具有重要的理论意义和应用价值。
中文关键词: 目标探测;特征提取;目标识别;机器学习;光学遥感
英文摘要: Though high spatial resolution remote sensing images provide more abundant and fine ground information, complete target often cannot be extracted automatically in complex scene, such as occluded, shadowed, noisy, cluttered, and have multi objects, by only local features (e.g.: spectral, texture, edge) derived from image data. Since we know the shapes of remote sensing targets, especially the shapes of artificial objects in advance, we propose this project to study targets extracting methods in complex scene using shape prior and image data. Based on the theories and methods in the fields of remote sensing, statistics, image registration, graph-based image segmentation and classification, aiming at buildings, roads, green fields in urban high resolution remote sensing images, statistical model for shape and object extracting methods with one shape template and multi shape templates using image data and shape prior similar to the object will be systematically studied in this project. This research has important academic and practical value in shape prior knowledge study and its remote sensing image application.
英文关键词: target detection;feature extraction;target identification;machine learning;optical sensing