融合级联CRFs和U-Net深度学习模型的遥感影像建筑物自动提取  

Automatic Building Extraction from Remote Sensing Images Based on Cascaded CRFs and the U-Net Deep Learning Model

作　　者：陈嘉浩 邢汉发 陈相龙 CHEN Jiahao;XING Hanfa;CHEN Xianglong(Guangdong Shengteng Geospatial Information Technology, Foshan 528200, China;School of Geography, South China Normal University, Guangzhou 510631, China;SCNU Qingyuan Institute of Science and Technology Innovation Co.,Ltd., Qingyuan 511517, China;Guangdong Normal University Weizhi Information Technology Co.,Ltd., Qingyuan 511517, China)

机构地区：[1]广东晟腾地信科技有限公司,佛山528200 [2]华南师范大学地理科学学院,广州510631 [3]华南师大(清远)科技创新研究院有限公司,清远511517 [4]广东师大维智信息科技有限公司,清远511517

出　　处：《华南师范大学学报（自然科学版）》2022年第1期70-78,共9页Journal of South China Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(41971406);广东省科技计划项目(2018B020207002)。

摘　　要：针对深度学习模型进行建筑物提取时存在的建筑物边缘模糊问题,将级联CRFs(全连接条件随机场)引入到U-Net深度模型中,提出了一种改进的U-Net模型(U-Net+级联CRFs),以用于遥感影像建筑物自动提取:构建级联CRFs并将其引入到U-Net模型的解码层中,从多层结构中学习边界信息,增强模型对建筑物边界的分割能力。并以广东省佛山市为研究区,利用U-Net+级联CRFs、U-Net+CRFs、U-Net、SVM模型进行建筑物提取实验。结果表明U-Net+级联CRFs模型可以有效识别建筑物边界信息,提高建筑物提取的精度:U-Net+级联CRFs模型在准确度、召回率、F1值和交并比4个指标上的均值分别达到了93.1%、87.5%、91.4%和85.1%,均优于U-Net+CRFs、U-Net、SVM模型。To address the problem of building edge ambiguity in deep learning models for building extraction,cascaded CRFs(fully connected conditional random fields)is introduced into the U-Net model and an improved U-Net model(U-Net+cascade CRFs)is proposed for automatic building extraction from remote sensing images.A cascaded CRFs model is constructed and introduced into the decoding layer of the U-Net model to learn the boundary information from the multi-layer structure and enhance the ability of the model to segment the building boundary.Taking Foshan City,Guangdong Province,as the research area,the U-Net+cascaded CRFs,U-Net+CRFs,U-Net and SVM models are used to carry out building extraction experiments.The results show that the proposed method can effectively identify building boundary information and improve the accuracy of building extraction:it can achieve 93.1%,87.5%,91.4%and 85.1%of the four indexes of accuracy,recall rate,F1 value and cross/combine ratio respectively,which are superior to those of U-Net+CRFs,U-Net and SVM models.

关 键 词：深度学习 建筑物提取 U-Net模型 全连接条件随机场 

分 类 号：P237[天文地球—摄影测量与遥感]