基于多源数据融合的城市土地利用精细识别方法  

Fine Urban Land Use Identification Based on Fusion of Multi-source Data

作　　者：李林超 钟良剑 苏庆 任璐 杜博文 LI Linchao;ZHONG Liangjian;SU Qing;REN Lu;DU Bowen(College of Civil and Transportation Engineering,Shenzhen University,Shenzhen 518060,China;School of Transportation Engineering,Chang’an University,Xi’an 710061,China;School of Computer Science and Engineering,Beihang University,Beijing 100191,China)

机构地区：[1]深圳大学土木与交通工程学院,广东深圳518060 [2]长安大学运输工程学院,陕西西安710061 [3]北京航空航天大学计算机学院,北京100191

出　　处：《西南交通大学学报》2025年第2期326-335,共10页Journal of Southwest Jiaotong University

基　　金：国家自然科学基金项目(52202402)。

摘　　要：我国土地利用类型复杂,为解决依靠单一遥感图像或POI(point of interest)数据而难以准确识别城市土地利用类型的困境,提出一种遥感图像和POI数据相结合的精细识别方法.首先,为精细识别城市地块功能,选取500 m栅格为研究单位;其次,提取POI数据并生成各类土地利用的核密度分布图,对遥感、POI图像数据进行数据预处理、数据切分、数据增强以提取有效信息;最后,将POI核密度分布图和高分遥感影像数据融合,以现状土地利用数据为标签,构建UNet++网络对城市地块分类,并运用CA算法对模型参数进行优化.以深圳市为实例开展实验,并在罗湖区和南山区进行迁移验证,结果表明:融合POI数据的城市土地利用精确识别模型平均精度为70.6%,比仅使用遥感数据识别模型精度高6.7%;使用CA算法后,模型精度提高1.5%;对模型进行迁移验证,模型平均精度为72.6%,表明模型具有较高的稳健性;此外,POI数据弥补了遥感影像仅涉及光谱、纹理和地物结构物理属性的不足,能较好识别商服用地、公共管理与公共服务用地,相较于单一数据识别模型精度分别高了7.5%、6.0%.Land use type in China is complex,and it is difficult to accurately identify urban land use type by relying on a single remote sensing image or point of interest(POI)data.To address this issue,a fine identification method combining remote sensing images and POI data was proposed.Firstly,to finely identify urban land parcel functions,a 500-meter grid was selected as the research unit;secondly,POI data were extracted,and kernel density distribution maps of various land uses were generated.Data preprocessing,data segmentation,and data enhancement were performed on remote sensing and POI image data to extract effective information.Finally,the POI kernel density distribution map and high-resolution remote sensing image data were fused together,and the current land use data was used as the label to construct the UNet++network to classify urban land parcels.The model parameters were optimized using the cosine annealing(CA)algorithm,and the proposed method was tested in Shenzhen City.Migration verification was carried out in Luohu District and Nanshan District.The results show that the average accuracy of the urban land use identification model fused with POI data is 70.6%,which is 6.7%higher than that of the identification model using only remote sensing data;after using the CA algorithm,the model accuracy is increased by 1.5%.The migration verification of the model is carried out,and the average accuracy of the model is 72.6%.This shows that the model is robust.In addition,POI data makes up for the shortcomings of remote sensing images that only involve spectrum,texture,and physical attributes of ground structures,and it can better identify commercial land and public management and service land.The accuracy is 7.5%and 6.0%higher than that of a single data identification model.

关 键 词：交通规划 土地利用 深度学习 多源数据 

分 类 号：U491[交通运输工程—交通运输规划与管理]