基于偏移阴影分析的高分辨率可见光影像建筑物自动提取

发布时间：2018-03-21 18:42

本文选题：遥感　切入点：高分辨率可见光遥感影像　出处：《光学学报》2017年04期 　论文类型：期刊论文

【摘要】：为了提高建筑物提取的自动化程度和精度,提出了一种以分割-分类-优化为主线、利用偏移阴影分析的建筑物全自动提取方法。首先,采用面向对象的多尺度分割方法进行影像初分割;然后,结合支持向量机(SVM)分类,将分割结果分为阴影、植被、建筑物、裸地四大类并提取初始结果;最后,利用相交边界阴影比率准确地验证了建筑物的存在,剔除了无阴影的非建筑物干扰,获取了最终结果。大量的实验结果验证了该方法的有效性,自动化程度得到明显提高。该方法完整度达到85%以上,正确率和综合分数F1均达到90%以上,且仅需要可见光波段影像数据,适用范围广。
[Abstract]:In order to improve the degree of automation and accuracy of building extraction, presents a segmentation classification optimization as the main line, using the offset shadow analysis of buildings automatic extraction method. Firstly, using object-oriented multi-scale segmentation method for image segmentation; then, combined with support vector machine (SVM) classification, segmentation results as the shadow, vegetation, bare land and buildings, four kinds of initial extraction results; finally, using intersecting boundary and shadowing ratio accurately verify the existence of the building, removing the interference of the buildings without the shadow, to obtain the final results. Experimental results verify the validity of the method, the degree of automation has been improved obviously. The integrity of the above 85%, the correct rate and comprehensive fraction of F1 reached more than 90%, and only need the visible band image data, a wide range of applications.

【作者单位】：长江大学地球科学学院;长江水利委员会长江科学院;天津市测绘院;
【基金】：国家自然科学基金(41671450,41371343) 地理国情监测国家测绘地理信息局重点实验室开放基金(2016NGCM07)
【分类号】：P237;TP751

【相似文献】