融合Sentinel-1/2深层特征与支持向量机的湿地制图方法  

Wetland mapping method integrating deep features and SVM using Sentinel-1/2

作　　者：王子璇 曹婧淼 舒菲娅 胥瀚文 牛玉芬 赵金奇 WANG Zixuan;CAO Jingmiao;SHU Feiya;XU Hanwen;NIU Yufen;ZHAO Jinqi(School of Environment Science and Spatial Informatics,China University of Mining and Technology;Key Laboratory of National Geographic Census and Monitoring,Ministry of Natural Resources;Observation and Research Station of Jiangsu Jiawang Resource Exhausted Mining Area Land Estoration and Ecological Succession,Ministry of Education,China University of Mining and Technology;Hebei University of Engineering)

机构地区：[1]中国矿业大学环境与测绘学院,江苏徐州221116 [2]自然资源部地理国情监测重点实验室,武汉430072 [3]江苏贾汪资源枯竭矿区土地修复与生态演替教育部野外科学观测研究站,江苏徐州221116 [4]河北工程大学,河北邯郸056038

出　　处：《测绘科学》2025年第1期114-124,共11页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(42307255,41901286);河北省自然科学基金项目(D2023402033);江苏省双创博士资助项目(JSSCBS20221531);自然资源部地理国情监测重点实验室开放基金项目(2023NGCM12);江苏省科技副总项目(FZ20240048)。

摘　　要：针对传统机器学习湿地制图方法存在地物特征表达不足的问题,提出了一种结合卷积神经网络预训练和支持向量机的湿地制图新方法。利用Sentinel-1和Sentinel-2数据,构建湿地多源观测数据集;利用CNN对少量真实样本进行预训练,构建预训练模型,对多源观测数据集进行深层特征提取;将提取的深层特征输入至SVM模型,从而获取湿地地物制图结果。选择黄河三角洲湿地为研究区验证本文方法的有效性,改进方法在湿地分类结果达到最优,OA达到88.1%,Kappa系数为0.859。实验结果表明:融合Sentinel-1和Sentinel-2影像能够有效提升湿地地物分类的精度;相较于其它分类器,深层特征对SVM分类器的提升最为明显。To address the limitation of wetland land-cover features using machine learning methods,a novel wetland mapping approach integrating pre-trained convolutional neural networks(CNN)and support vector machine(SVM)is proposed in this paper.Firstly,optical and synthetic aperture radar(SAR)images from Sentinel-1 and Sentinel-2 are used to construct a multi-source wetland dataset.In addition,a CNN model is pre-trained on a small set of labeled samples to enable deep feature extraction.Furthermore,the wetland mapping result is obtained using SVM with deep features.The Yellow River Delta wetland is chosen as the study area to evaluate the effectiveness of the proposed method.The proposed method achieves superior classification performance,with an OA of 88.1%and a Kappa coefficient of 0.859.The experimental results demonstrate that:the integration of Sentinel-1 and Sentinel-2 data significantly improves the accuracy of wetland mapping,compared with the other classifiers,deep features by pre-trained is the most suitable for SVM classifier.

关 键 词：湿地制图 多源遥感 卷积神经网络 支持向量机 

分 类 号：P237[天文地球—摄影测量与遥感]