随机森林和图神经网络支持下的河系自动分级与选取方法

doi:10.11947/j.AGCS.2025.20240385

测绘学报 ›› 2025, Vol. 54 ›› Issue (9): 1697-1711.doi: 10.11947/j.AGCS.2025.20240385

随机森林和图神经网络支持下的河系自动分级与选取方法

张付兵¹^,²(), 孙群¹(), 徐青¹^,³^,⁴, 马京振¹^,⁵, 黄文君¹, 陈若虚¹

^1.信息工程大学地理空间信息学院，河南　郑州　450052
^2.陆军工程大学爆炸冲击防灾减灾全国重点实验室，江苏　南京　210007
^3.智慧中原地理信息技术河南省协同创新中心，河南　郑州　450052
^4.时空感知与智能处理自然资源部重点实验室，河南　郑州　450052
^5.61540部队，陕西　西安　710054

收稿日期:2024-09-16 修回日期:2025-09-15 出版日期:2025-10-10 发布日期:2025-10-10
通讯作者: 孙群 E-mail:zhangfbing@163.com;13503712102@163.com
作者简介:张付兵（1997—），男，博士，讲师，主要研究方向为空间数据挖掘与多尺度表达。E-mail：zhangfbing@163.com
基金资助:
国家自然科学基金(42101454);中国博士后科学基金(2024M764345)

An automatic river classification and selection method supported by random forest and graph neural network

Fubing ZHANG¹^,²(), Qun SUN¹(), Qing XU¹^,³^,⁴, Jingzhen MA¹^,⁵, Wenjun HUANG¹, Ruoxu CHEN¹

^1.Institute of Geospatial Information, University of Information Engineering, Zhengzhou 450052, China
^2.State Key Laboratory of Disaster Prevention & Mitigation of Explosion & Impact, Army Engineering University of PLA, Nanjing 210007, China
^3.Collaborative Innovation Center of Geo-information Technology for Smart Central Plains, Zhengzhou 450052, China
^4.Key Laboratory of Spatiotemporal Perception and Intelligent processing, Ministry of Natural Resources, Zhengzhou 450052, China
^5.Troops 61540, Xi'an 710054, China

Received:2024-09-16 Revised:2025-09-15 Online:2025-10-10 Published:2025-10-10
Contact: Qun SUN E-mail:zhangfbing@163.com;13503712102@163.com
About author:ZHANG Fubing (1997—), male, PhD, lecturer, majors in spatial data mining and multi-scale representation. E-mail: zhangfbing@163.com
Supported by:
The National Natural Science Foundation of China(42101454);China Postdoctoral Science Foundation(2024M764345)

摘要/Abstract

摘要：

随着深度学习方法在地图制图领域的广泛应用，利用图神经网络解决非结构化矢量地图数据综合问题成为当前研究的热点。针对现有河系分级方法主要从局部结构出发，应用传统机器学习方法未考虑相邻河段之间的局部相关性，且基于图神经网络的河系选取方法仅考虑河段之间拓扑邻接关系的不足，本文提出了一种随机森林和图神经网络支持下的河系自动分级与选取方法。首先，通过融入河系分级知识并通过特征局部标准化处理，利用随机森林算法实现河系自动分级；然后，引入河段之间拓扑邻接关系和空间邻近关系构建双支图神经网络选取模型，通过监督学习实现河段的选取分类；最后，通过兼顾层次分级的连通性保持策略获取河系选取结果。试验结果表明，相较Min-Max标准化和Z-Score标准化，本文河系自动分级精度分别提升了11.42个百分点和12.39个百分点，分级效果更好；相较已有基于图神经网络的河系选取方法，选取精度提升了2.2个百分点，并且与标签数据之间的差异更小。

关键词: 河系选取, 自动分级, 随机森林, 图神经网络, 特征局部标准化

Abstract:

Nowadays, with the widespread use of deep learning methods in the field of cartography, the use of graph neural networks (GNNs) to solve the generalization problems of unstructured vector map data has become a research hotspot. A river network automatic classificationand selection method supported by random forests and GNN is proposed to address the shortcomings of the existing methods, which mainly start from local structures and do not consider the local correlation of adjacent river segments when applying machine learning methods to classify rivers, and only consider the topological relationship of river segments in the process of river network selection. Firstly, incorporating knowledge of river network classification and local standardization of features, the random forest algorithm is used to automatically classify river network. Then, a dual branch GNN selection model is constructed by integrating topological connections and spatial proximity relationships between river segments, and the selected classification of river segments is achieved through supervised learning. Finally, the river network selection results are obtained by adopting a connectivity preservation strategy that considers hierarchical levels. The experimental results show that compared with standardization ways of Min-Max and Z-Score, the automatic classification accuracy of the river network has been improved by 11.42 percentage points and 12.39 percentage points respectively, and the classification effect is better. Compared with existing river network selection methods based on graph neural networks, the selection accuracy has been improved by 2.2 percentage points, and closer to the labeled data.

Key words: river network selection, automatic classification, random forest, graph neural network, feature local standardization

中图分类号:

P208

张付兵, 孙群, 徐青, 马京振, 黄文君, 陈若虚. 随机森林和图神经网络支持下的河系自动分级与选取方法[J]. 测绘学报, 2025, 54(9): 1697-1711.

Fubing ZHANG, Qun SUN, Qing XU, Jingzhen MA, Wenjun HUANG, Ruoxu CHEN. An automatic river classification and selection method supported by random forest and graph neural network[J]. Acta Geodaetica et Cartographica Sinica, 2025, 54(9): 1697-1711.

图/表 24

图1

图2

图3

图4

图5

图6

图7

图8

表1

表2

表3

图9

表4

表5

图10

表6

图11

表7

图12

表8

表9

图13

图14

图15

参考文献 27

[1]	谭笑, 武芳, 黄琦, 等. 主流识别的多准则决策模型及其在河系结构化中的应用[J]. 测绘学报, 2005, 34(2): 154-160.
	TAN Xiao, WU Fang, HUANG Qi, et al. A multi-criteria decision model for identifying master river and its application in river system construction[J]. Acta Geodaetica et Cartographic Sinica, 2005, 34(2): 154-160.
[2]	YU Huafei, AI Tinghua, YANG Min, et al. A recognition method for drainage patterns using a graph convolutional network[J]. International Journal of Applied Earth Observation and Geoinformation, 2022, 107: 102696. DOI: . doi: 10.1016/j.jag.2022.102696
[3]	YU Huafei, AI Tinghua, YANG Min, et al. A graph autoencoder network to measure the geometric similarity of drainage networks in scaling transformation[J]. International Journal of Digital Earth, 2023, 16(1): 1828-1852. DOI: . doi: 10.1080/17538947.2023.2212920
[4]	WANG Wenning, YAN Haowen, LU Xiaomin, et al. Drainage pattern recognition method considering local basin shape based on graph neural network[J]. International Journal of Digital Earth, 2023, 16(1): 593-619. DOI: . doi: 10.1080/17538947.2023.2172224
[5]	YU Huafei, AI Tinghua, YANG Min, et al. Integrating domain knowledge and graph convolutional neural networks to support river network selection[J]. Transactions in GIS, 2023, 27(7): 1898-1927. DOI: . doi: 10.1111/tgis.13104
[6]	WANG Di, QIAN Haizhong. Graph neural network method for the intelligent selection of river system[J]. Geocarto International, 2023, 38: 2252762. DOI: . doi: 10.1080/10106049.2023.2252762
[7]	段佩祥, 钱海忠, 何海威, 等. 案例支撑下的朴素贝叶斯树状河系自动分级方法[J]. 测绘学报, 2019, 48(8): 975-984. DOI: . doi: 10.11947/j.AGCS.2019.20180370
	DUAN Peixiang, QIAN Haizhong, HE Haiwei, et al. Naive Bayes-based automatic classification method of tree-like river network supported by cases[J]. Acta Geodaetica et Cartographica Sinica, 2019, 48(8): 975-984. DOI: . doi: 10.11947/j.AGCS.2019.20180370
[8]	HORTON R E. Erosional development of streams and their drainage basins; hydrophysical approach to quantitative morphology[J]. Geological Society of America Bulletin, 1945, 56(3): 275-370.
[9]	STRAHLER A N. Quantitative analysis of watershed geomorphology[J]. Eos, Transactions American Geophysical Union, 1957, 38(6): 913-920. DOI: . doi: 10.1029/TR038i006p00913
[10]	SHREVE R L. Statistical law of stream numbers[J]. The Journal of Geology, 1966, 74(1): 17-37.
[11]	张园玉, 李霖, 金玉平, 等. 基于图论的树状河系结构化绘制模型研究[J]. 武汉大学学报(信息科学版), 2004, 29(6): 537-539, 543.
	ZHANG Yuanyu, LI Lin, JIN Yuping, et al. Structured design of dendritic river networks based on graph[J]. Geomatics and Information Science of Wuhan University, 2004, 29(6): 537-539, 543.
[12]	翟仁键, 薛本新. 面向自动综合的河系结构化模型研究[J]. 测绘科学技术学报, 2007, 24(4): 294-298, 302.
	ZHAI Renjian, XUE Benxin. A structural river network data model for automated river generalization[J]. Journal of Zhengzhou Institute of Surveying and Mapping, 2007, 24(4): 294-298, 302.
[13]	郭庆胜, 黄远林. 树状河系主流的自动推理[J]. 武汉大学学报(信息科学版), 2008, 33(9): 978-981.
	GUO Qingsheng, HUANG Yuanlin. Automatic reasoning on main streams of tree river networks[J]. Geomatics and Information Science of Wuhan University, 2008, 33(9): 978-981.
[14]	PAIVA J, EGENHOFER M J, FRANK A U. Spatial reasoning about flow directions: towards an ontology for river networks[C]//Proceedings of 1992 XV International Congress for Photogrammetry and Remote Sensing. [S. l.]: ISPRS, 1992: 224-318.
[15]	李成名, 殷勇, 吴伟, 等. Stroke特征约束的树状河系层次关系构建及简化方法[J]. 测绘学报, 2018, 47(4): 537-546. DOI: . doi: 10.11947/i.AGCS.2018.20170141
	LI Chengming, YIN Yong, WU Wei, et al. Method of tree-like river networks hierarchical relation establishing and generalization considering Stroke properties[J]. Acta Geodaetica et Cartographica Sinica, 2018, 47(4): 537-546. DOI: . doi: 10.11947/i.AGCS.2018.20170141
[16]	THOMSON R C, RICHARDSON D E. The “good continuation” principle of perceptual organization applied to the generalization of road networks[C]//Proceedings of the ICA 19th international cartographic conference. Ottawa: IEEE, 1999: 1215-1223.
[17]	翟仁健, 武芳, 邓红艳, 等. 基于遗传多目标优化的河流自动选取模型[J]. 中国矿业大学学报, 2006, 35(3): 403-408.
	ZHAI Renjian, WU Fang, DENG Hongyan, et al. Automated elimination of river based on multi-objective optimization using genetic algorithm[J]. Journal of China University of Mining & Technology, 2006, 35(3): 403-408.
[18]	艾廷华, 刘耀林, 黄亚锋. 河网汇水区域的层次化剖分与地图综合[J]. 测绘学报, 2007, 36(2): 231-236, 243.
	AI Tinghua, LIU Yaolin, HUANG Yafeng. The hierarchical watershed partitioning and generalization of river network[J]. Acta Geodaetica et Cartographica Sinica, 2007, 36(2): 231-236, 243.
[19]	邵黎霞, 何宗宜, 艾自兴, 等. 基于BP神经网络的河系自动综合研究[J]. 武汉大学学报(信息科学版), 2004, 29(6): 555-557.
	SHAO Lixia, HE Zongyi, AI Zixing, et al. Automatic generalization of river network based on BP neural network techniques[J]. Geomatics and Information Science of Wuhan University, 2004, 29(6): 555-557.
[20]	SEN A, GOKGOZ T, SESTER M. Model generalization of two different drainage patterns by self-organizing maps[J]. Cartography and Geographic Information Science, 2014, 41(2): 151-165.
[21]	SEN A, GOKGOZ T. An experimental approach for selection/elimination in stream network generalization using support vector machines[J]. Geocarto International, 2015, 30(3): 311-329.
[22]	YAN Chaode, LIU Xiao, BOOTA M W, et al. A comparison of machine learning methods applied to the automated selection of river networks[J]. The Cartographic Journal, 2022, 59(3): 187-202.
[23]	NIEPERT M, AHMED M, KUTZKOV K. Learning convolutional neural networks for graphs[C]//Proceedings of the 33rd International Conference on Machine Learning. New York: IEEE, 2016: 2014-2023.
[24]	KIPF T N, WELLING M. Semi-supervised classification with graph convolutional networks[EB/OL]. [2025-01-25]. https//arxiv.org/abs1609.
[25]	HAMILTON W, YING Z. LESKOVEC J. Inductive representation learning on large graphs[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. California: IEEE, 2017: 1025-1035. DOI: . doi: 10.48550/arXiv.1706.02216
[26]	BREIMAN L. Random forests[J]. Machine Learning, 2001, 45(1): 5-32.
[27]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780.

选取单元河段数统计	训练集（700）		测试集（71）
选取单元河段数统计	1∶24 000	1∶250 000	1∶24 000	1∶250 000
平均值	627.7	212.2	549.0	191.9
最小值	85.0	37.0	137.0	49.0
最大值	1 684.0	556.0	1 476.0	460.0

河系编号	训练集			河系编号	测试集
河系编号	主流数	支流数	图示	河系编号	主流数	支流数	图示
河系1	133	132		河系6	450	454
河系2	371	372		河系7	550	551
河系3	205	205		河系8	478	480
河系4	391	391		河系9	378	377
河系5	341	341		河系10	160	159

标准化方式	测试集分类准确率	主支流识别准确率
Min-Max标准化	84.42	90.34
Z-Score标准化	83.45	89.05
特征局部标准化	95.84	97.03

标准化方式	特征重要性
标准化方式	RLen	URL	WArea	URA	BAngle	URN
Min-Max标准化	0.066	0.161	0.085	0.145	0.350	0.192
Z-Score标准化	0.072	0.154	0.092	0.134	0.364	0.185
特征局部标准化	0.031	0.325	0.043	0.330	0.142	0.130

机器学习算法	测试集分类准确率	主支流识别准确率
高斯分布朴素贝叶斯	92.62	94.60
决策树	93.88	93.85
随机森林	95.84	97.03

随机森林和图神经网络支持下的河系自动分级与选取方法

An automatic river classification and selection method supported by random forest and graph neural network

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 24

参考文献 27

相关文章 13

编辑推荐

Metrics

本文评价

分级方式	训练集	测试集
分级方式	验证准确率	准确率	精度	召回率	F₁值
Hotron	87.50	86.16	86.09	86.16	86.11
Strahler	88.69	88.29	88.29	88.09	88.07
Shreve	89.87	89.36	89.38	89.36	89.18
Stroke	82.59	81.89	81.62	81.89	81.63
RFCoding	89.83	89.56	89.49	89.56	89.49

特征组合	训练集	测试集
特征组合	验证准确率	准确率	精度	召回率	F₁
+Hotron	91.28	90.86	90.89	90.89	90.77
+Strahler	91.10	90.77	90.72	90.77	90.71
+Shreve	90.96	90.83	90.81	90.84	90.74
+Stroke	91.11	90.79	90.76	90.79	90.69
+RFCoding	92.01	91.80	91.76	91.80	91.77

方法	精度	召回率	F₁
决策树	90.05	89.66	90.25
随机森林	90.23	89.82	90.06
GAT法^[5]	90.70	90.40	90.40
GraphSAGE法^[5]	91.20	90.60	90.70
本文单支选取模型	92.36	92.39	92.33
本文双支选取模型	92.63	92.66	92.62

初始数据	方法	精度	选取标签的精度	选取标签的召回率	选取标签的F₁值
单支选取模型结果	连通处理前	92.36	91.72	85.98	88.76
	最长路径法	93.00	86.90	93.36	90.01
	层次连通法	93.32	91.46	89.28	90.36
双支选取模型结果	连通处理前	92.63	91.50	87.10	89.25
	最长路径法	93.22	87.03	93.83	90.30
	层次连通法	93.40	91.15	89.89	90.52

[1]	王亚青, 王中辉. 异构图卷积网络支持下的河系自动选取方法[J]. 测绘学报, 2025, 54(7): 1332-1345.
[2]	黄哲琨, 钱海忠, 蔡中祥, 王骁, 王俊威, 孔令辉. 基于图神经网络的多尺度网状河系分类匹配方法[J]. 测绘学报, 2025, 54(2): 371-384.
[3]	唐佳怡, 童晓冲, 邱春平, 雷亚现, 雷毅, 宋好帅. 基于场景图的遥感场景检索方法[J]. 测绘学报, 2025, 54(1): 123-135.
[4]	陈超, 梁锦涛, 杨刚, 孙伟伟, 龚绍军, 王建强. 面向土地覆盖精准分类的遥感特征参数优选方法[J]. 测绘学报, 2024, 53(7): 1401-1416.
[5]	孔博, 艾廷华, 杨敏, 吴昊, 余华飞, 肖天元. 等高线形态知识与图神经网络联合作用下的黄土地貌类型识别[J]. 测绘学报, 2024, 53(6): 1154-1164.
[6]	陈占龙, 鲁谢春, 徐永洋. 基于图顶点深度聚类的建筑物合并方法[J]. 测绘学报, 2024, 53(4): 736-749.
[7]	王蓝星, 王群明, 童小华. 融合多光谱影像的高光谱影像厚云去除方法[J]. 测绘学报, 2022, 51(4): 612-621.
[8]	王猛, 张新长, 王家耀, 孙颖, 箭鸽, 潘翠红. 结合随机森林面向对象的森林资源分类[J]. 测绘学报, 2020, 49(2): 235-244.
[9]	段佩祥, 钱海忠, 何海威, 谢丽敏, 罗登瀚. 案例支撑下的朴素贝叶斯树状河系自动分级方法[J]. 测绘学报, 2019, 48(8): 975-984.
[10]	熊艳, 高仁强, 徐战亚. 机载LiDAR点云数据降维与分类的随机森林方法[J]. 测绘学报, 2018, 47(4): 508-518.
[11]	杨俊涛, 康志忠. 多尺度特征和马尔可夫随机场模型的电力线场景点云分类法[J]. 测绘学报, 2018, 47(2): 188-197.
[12]	张志强, 张新长, 辛秦川, 杨晓羚. 结合像元级和目标级的高分辨率遥感影像建筑物变化检测[J]. 测绘学报, 2018, 47(1): 102-112.
[13]	冯文卿, 眭海刚, 涂继辉, 孙开敏, 黄伟明. 高分辨率遥感影像的随机森林变化检测方法[J]. 测绘学报, 2017, 46(11): 1880-1890.