光学学报, 2017, 37 (4): 0428002, 网络出版: 2017-04-10
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
Automatic Building Extraction from High Resolution Visible Images Based on Shifted Shadow Analysis
遥感 高分辨率可见光遥感影像 建筑物自动提取 影像分割分类 分类样本自动获取 建筑物验证 remote sensing high resolution visible remote sensing images automatic building extraction image segmentation and classification automatic sample extraction of classified objects building verification
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 automation level and the precision of building extraction, an automatic building extraction method based on shifted shadow analysis is proposed. It is guided by the principal line of segmentation-classification-optimization. The object oriented multi-resolution segmentation method is adopted to perform the initial image segmentation. The segmentation results are classified by the support vector machine (SVM) classifier into four categories, i.e., shadow, vegetation, building and bare land. The initial results are extracted. The shadow rate on the intersection boundary is designed to accurately validate the existence of buildings and remove the disruptions of non-buildings without shadows, and the final results are obtained. The large amount of experimental results validate that the proposed method is very effective, and the automation level is significantly improved. The completeness is more than 85%. The correctness and the F1-score can both reach more than 90%.The proposed method only needs data from images in the visible band and has a wide application range.
高贤君, 郑学东, 刘子潇, 杨元维. 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[J]. 光学学报, 2017, 37(4): 0428002. Gao Xianjun, Zheng Xuedong, Liu Zixiao, Yang Yuanwei. Automatic Building Extraction from High Resolution Visible Images Based on Shifted Shadow Analysis[J]. Acta Optica Sinica, 2017, 37(4): 0428002.