多特征多尺度相结合的高分辨率遥感影像建筑物提取

在高分辨率遥感影像中，建筑物通常表现为多尺度形态，且存在同谱异物和同物异谱现象。因此，本文提出了一种综合利用光谱特征、形状特征和纹理特征，并结合多尺度分割的建筑物分级提取方法。该方法首先对遥感影像进行形态学建筑物指数（MBI）计算，而后对其特征影像进行阈值分割，并借助形状特征参数实现建筑物初提取；然后引入面向对象思想完成遥感影像多尺度分割，并利用纹理特征实现单一尺度的建筑物对象识别；最后借助多尺度融合思想完成建筑物后提取。利用本文方法对冲绳某地区影像进行了建筑物提取试验。试验结果表明，该方法的识别查准率和查全率在对象级和像素级两方面均取得较高精度。

Building Extraction from High Resolution Remote Sensing Imagery with Multi-feature and Multi-scale

Buildings vary in different shapes and sizes in high resolution remote sensing imageries , and the phenomenon that large within-class spectral variations and between-class spectral confusions also exist .In this paper , a method based on spectral feature , shape feature, texture feature and multilevel segmentation is proposed .Firstly, the original extraction is carried out by calculating the MBI, threshold segmentation and shape features .Then, object-oriented analysis is used for multi-scale segmentation and texture feature is used for building recognition in single scale .Finally, multi-scale fusion is used for the ultimate extraction .The presented method is evaluated with an image of Okinawa , Japan.The experiments show that the proposed building extraction algorithm can provide satisfactory precision ratio with a high level of recall ratio .