基于特征优选的GF-6WFV影像湿地信息提取  

Wetland Information Extraction from GF-6 WFV Multispectral Imagery Based on Feature Optimization

作　　者：黄冰鑫 徐佳[1] 张晓同 陈成[2] HUANG Bingxin;XU Jia;ZHANG Xiaotong;CHEN Cheng(School of Earth Sciences and Engineering,Hohai University,Nanjing 211100,China;Jiangsu Province Surveying&Mapping Engineering Institute,Nanjing 210013,China)

机构地区：[1]河海大学地球科学与工程学院,江苏南京211100 [2]江苏省测绘工程院,江苏南京210013

出　　处：《地理空间信息》2024年第4期39-44,共6页Geospatial Information

基　　金：自然资源部国土卫星遥感应用重点实验室开放基金资助项目(KLSMNR-K202209);中央高校基本科研业务费专项资金资助项目(B220202052);江苏省自然资源发展专项资金资助项目(JSZRHYKJ202101)。

摘　　要：以洪泽湖湿地为研究对象,构建一种基于特征优选的湿地面向对象分类方法,利用高分六号宽幅多光谱(GF-6 WFV)影像进行湿地信息提取。首先对GF-6 WFV数据进行预处理和多尺度分割;然后提取光谱、植被指数、水体指数、红边指数和纹理特征,利用基于平均准确度下降法(mean decrease accuracy,MDA)的递归排除算法生成最优特征集;最后,基于最优特征集进行湿地分类,通过对比4种特征优选算法发现基于MDA的递归排除算法能更有效地进行特征变量选择,使用6种方案开展湿地分类实验。结果表明GF-6 WFV影像的红边波段和红边指数在湿地分类中具有重要作用,利用优选特征集的分类精度最高,为87.43%,比未进行特征优选的分类精度提高了1.55%。研究成果将为GF-6 WFV影像在湿地监测方面提供技术参考。Taking the Hongze Lake wetland as the study area,we proposed an object-oriented wetland classification method based on feature optimization,and used GF-6 WFV multispectral imagery to obtain wetland information.Firstly,GF-6 WFV data was preprocessed and multi-scale segmentation.Then,we extracted the characteristic variables,including spectral feature,vegetation index,water index,red edge index and texture feature from GF-6 WFV images,and obtained the optimal feature set by using a recursive elimination algorithm with mean decrease accuracy(MDA).Finally,we used the object-oriented classification method to extract the wetland information based on the optimal feature combination.Comparing with four different feature optimization algorithms,we found that the recursive elimination algorithm with MDA can select feature variables more effectively.We used different feature combinations in the Hongze Lake wetland to carry out six classification experiments.The results indicated that the overall accuracy of proposed method up to 87.43%is the highest,improving 1.55%compared with the method without feature optimization.Additionally,the two red-edge bands and red-edge indexes of GF-6 WFV data play an important role in wetland classification.The research results will provide technical reference for the application of GF-6 WFV multispectral images in wetland monitoring.

关 键 词：GF-6 湿地分类 特征优选 随机森林 红边波段 

分 类 号：P237[天文地球—摄影测量与遥感]