基于Sentinel-2A的农田包围型村落提取

FARMLAND SURROUNDED VILLAGES EXTRACTION BASED ON SENTINEL-2A DATA

遥感数据源在光谱、空间分辨率上不断提高,越来越有利于村落的精确提取,但如何有效利用影像特征,寻找简单便捷的方法来实现农田包围型村落的准确快速提取是一个不断需要探索的问题。本研究以Sentinel-2A卫星影像为数据源,利用光谱指数密度分割法与光谱波段分类法对内蒙古五原县郊区的村落进行提取。结果显示:NDVI,NDBI,NDBI-NDVI.(NDBI-NDVI)/(NDBI + NDVI)4个光谱指数提取村落的总体精度介于93.903 4%-96.476 6%之间,NDVI提取精度最低,NDBI提取精度最高;光谱波段分类法提取的精度介于94. 101 3%-98.753 0%之间,且利用蓝、绿光波段分类取得最高精度。研究结果表明,分类法和阈值法提取村落均可取得较高的精度,MLC-RFE法可有效获得最有利于村落提取的波段组合,但过程较繁琐、速度慢,阈值法精度略低,但较简单,计算速度快。

Remote sensing data sources are improving in spectral and spatial resolution, which is more and more conducive to the accurate extraction of villages, however, how to effectively use the image features to find a simple and convenient method to achieve accurate and rapid extraction of farmland surrounding villages is a problem that needs to be explored. In this study, Sentinel-2A satellite image is used as the data source, the villages in the suburbs of Wuyuan County in Inner Mongolia were extracted by spectral index density segmentation and spectral band classification. Results showed that: The overall accuracy of the 4 spectral indices NDVI、NDBI 、NDBI?NDVI and (NDBI-NDVI)/(NDBI + NDVI) for the extraction of villages ranged from 93. 903 4% to 96. 476 6%, the precision of NDVI extraction was the lowest, and the precision of NDBI extraction was the highest;the overall accuracy of spectral band classification is between 94. 101 3% and 98. 753 0%, and the highest accuracy is achieved by the classification using blue and green bands. The results showed that the classification method and threshold method both can obtain high accuracy, the MLC-RFE method can effectively track the feature combination which is most beneficial to the village extraction, however, the process is complex and slow, the density slice method is slightly lower, but it is simple and fast.