摘要
本文利用GEE(Google Earth Engine)平台中的多源遥感数据与随机森林分类器对廊坊城区黑臭水体进行了提取.研究结果表明:(1)光学遥感数据中的红光、绿光和蓝光波段可以准确反映水色信息,而雷达后续散射数据可以在一定程度上反映污水表面的几何信息,融合光学与微波的多源遥感特征可以更加准确地的提取黑臭水体信息.(2)随机森林分类器的训练精度达到96.7％,分类精度达到了 93.7％,可以基本满足对城市黑臭水体监测的要求.(3)在非黑臭水体的岸边、干涸区域、桥梁等会出现分类错误,这些分类错误是由多源遥感数据的在黑臭和非黑臭水体属性特征空间上的重叠造成的,加大样本点的数量与更多特征属性的选择将有助于提高这部分区域的分类精度.(4)免费的数据与算法可以为环保相关部门低成本定期监测城区黑臭水体提供切实可行、廉价的解决方案.
Abstract
This paper uses multi-source remote sensing data and random forest classifier in the GEE,Google Earth Engine,platform to extract black odorous water bodies in Langfang City.The results show that the red,green and blue bands in the optical remote sensing data can accurately reflect the water color information,and the radar back scattering data can reflect the geometric information of the sewage surface to a certain extent.Multi-source remote sensing features can help extracting black odorous water information more accurately.Furthermore,the training accuracy of the random forest classifier reaches 96.7％,and the classification accuracy reaches 93.7％,which can basically meet the requirements for monitoring urban black odorous water bodies.Besides,classification errors will occur on the shores of non-black odorous water bodies,dry areas,bridges,etc.These classification errors are caused by the overlap in the feature space of multi-source remote sensing data of black odorous water bodies and non-black odorous water bodies.The number of large sample points and the selection of more feature attributes will help to improve the classification accuracy of this area.Hopefully,free data and algorithms can provide practical solutions for environmental protection departments to regularly monitor black odorous water bodies in urban areas at low cost.
基金项目
河北省高等学校科学技术研究青年基金(QN2020429)
北华航天工业学院青年基金(KY-2020-05)
河北省教育厅重点项目(ZD2022089)
NETL
NSTL
万方数据