顾及多分辨率特征的复合字典城中村识别方法  

Identification method of urban villages with improved composite dictionary considering multi-resolution features

作　　者：邢若芸 冉树浩 高贤君[1,2,3] 杨元维 方军 XING Ruoyun;RAN Shuhao;GAO Xianjun;YANG Yuanwei;FANG Jun(School of Geosciences,Yangtze University,Wuhan 430100,China;Hunan Provincial Key Laboratory of Geo-information Engineering in Surveying,Mapping and Remote Sensing,Hunan University of Science and Technology,Xiangtan 411201,China;National-local Joint Engineering Laboratory of Geo-spatial Information Technology,Hunan University of Science and Technology,Xiangtan 411201,China;Beijing Key Laboratory of Urban Spatial Information Engineering,Beijing Institute of Surveying and Mapping,Beijing 100045,China)

机构地区：[1]长江大学地球科学学院,湖北武汉430100 [2]湖南科技大学测绘遥感信息工程湖南省重点实验室,湖南湘潭411201 [3]湖南科技大学地理空间信息技术国家地方联合工程实验室,湖南湘潭411201 [4]城市空间信息工程北京市重点实验室,北京100045

出　　处：《测绘通报》2023年第4期41-48,共8页Bulletin of Surveying and Mapping

基　　金：湖南科技大学测绘遥感信息工程湖南省重点实验室开放基金(E22205);自然资源部环鄱阳湖区域矿山环境监测与治理重点实验室开放基金(MEMI-2021-2022-08);城市轨道交通数字化建设与测评技术国家工程实验室开放课题基金(2021ZH02);城市空间信息工程北京市重点实验室经费(20210205)。

摘　　要：城中村作为一种特殊的城市聚落类型,对其进行精确有效的监控识别有助于实现城乡协调发展、优化城乡生态环境。现有面向对象的城中村识别方法通常需要大量样本数据,导致训练成本较高,数据更新效率偏低。针对以上问题,本文提出了顾及多分辨率特征的复合字典城中村识别方法。首先通过密集格网采样提取尺度不变特征转换(SIFT)全局特征,并与多分辨率颜色矢量角直方图特征融合,形成视觉词典;然后将影像表示为视觉词频率直方图;最后使用随机森林分类器进行分类,以实现场景尺度的城中村识别。以高分二号影像为测试数据对该方法进行验证,结果表明,其总体精度达90.08%,Kappa系数达80.16%,相较于加速稳健特征(SURF)、SIFT、VGG16、ResNet50,总体精度分别高出8.99%、3.51%、4.78%、2.28%。Urban village is a special type of urban settlement,and accurate and effective monitoring and identification of urban village is conducive to achieving coordinated development of urban and rural areas and optimizing the ecological environment.Existing object-oriented urban village identification methods usually require a large amount of sample data,resulting in high training cost and low data update efficiency.In order to solve the problems,a composite dictionary urban village identification method considering multi-resolution characteristics is proposed.Firstly,we use dense grid sampling to extract SIFT global features,and fuse them with multi-resolution color vector angular histogram features to form a visual dictionary.Then we use the image representation as a visual word frequency histogram.Finally,the random forest classifier is classified to realize the identification of urban villages at scene scale.The overall accuracy of the proposed method reaches 90.08%and the Kappa coefficient reaches 80.16%.It is 8.99%,3.51%,4.78%and 2.28%higher than that of SURF,SIFT,VGG16 and ResNet50,respectively.

关 键 词：城中村识别 高分辨率遥感影像 复合字典 多分辨率颜色特征 直方图特征融合 

分 类 号：P237[天文地球—摄影测量与遥感]