一种传统民居遥感提取方法

Traditional residence extraction from high-resolution imagery based on morphological building index and texture

针对基于地物光谱统计特征的建筑物提取方法由于存在较大的同物异谱现象导致提取结果不满足要求的问题,该文提出了一种基于形态学建筑物指数并顾及纹理特征的遥感提取方法。该方法综合考虑传统民居在高分辨率遥感影像上的光谱、形态和纹理特征,首先利用形态学建筑物指数法提取建筑,并使用最小矩形长宽比和像元个数区分道路和零星地物,而后利用Contourlet变换和谱直方图相似性计算进行纹理甄别,实现传统民居的遥感识别和提取。为了验证该方法,选取湖南省常宁市庙前镇中田村QuickBird影像进行试验,结果表明该方法能够获得较高精度的提取结果,整体精度为71.54%,影响提取精度的关键原因为损毁严重的建筑物光谱特征与目标图像纹理相差较大。

Remote sensing is one of potential technologies that can be used to obtain traditional resi- dence distribution from imagery, and extraction of traditional residence base remote sensing imagery is an important subject in human geography. A new method based on building morphological index and texture characters was proposed, and it combined features including spectral, morphology and texture. Based on this method, building pixels were firstly extracted using morphological building index algorithm~ then u- sing ratio between length and width and number of pixels of minimum boundary rectangle to get rid of road and partial pixels; finally, the contourlet transform algorithm and histogram of frequency spectrum were used to recognize the traditional residence pixels. In order to invalid this method, an experimental study was conducted using Quickbird imagery, whieh was located at Zhongtian village at Zhongtian town in Changning county of Hunan province. The result showed that accuracy of this method was satisfied with 71.54%, and the key factor affecting result accuracy was that omit of pixels which located at destroyed traditional residence.