面向微地图的手绘地图符号自动提取方法
余懿韬(2000— ),男,湖南常德人,硕士生,主要从事微地图研究。E-mail: 12222001@stu.lzjtu.edu.cn |
Copy editor: 蒋树芳
收稿日期: 2024-02-27
修回日期: 2024-04-01
网络出版日期: 2024-06-25
基金资助
甘肃省高等学校产业支撑计划项目(2022CYZC-30)
国家自然科学基金重大项目课题(42394063)
中国科学院数字地球重点实验室开放基金项目(2022LDE004)
Automatic Extraction Method of Point Symbols in Modern Hand-Drawn Maps for We-Map
Received date: 2024-02-27
Revised date: 2024-04-01
Online published: 2024-06-25
Supported by
The Industrial Support and Program Project of Universities in Gansu Province(2022CYZC-30)
The National Natural Science Foundation of China(42394063)
The Open Research Fund of Key Laboratory of Digital Earth Science, Chinese Academy of Sciences(2022LDE004)
微地图作为自媒体时代出现的新型地图,具有大众参与、个性化、快速传播等特征,然而现有微地图研究在点符号设计方面存在一定局限,难以完全满足大众个性化制图的需求。为解决这一问题,本文从微地图制作角度出发,选择手绘地图的通用地图符号作为研究对象,构建了一个包含多种类型和样式的手绘地图数据集。在现有研究的基础上,通过对比选择目标检测中常用的YOLOv5(You Only Look Once v5)系列模型,深入探索手绘地图中通用地图符号的自动提取方法,并采用YOLOv5-X模型进行手绘地图通用地图符号的提取。实验结果显示,该模型在手绘地图数据集上的point类别提取精确度、召回率和F1得分分别达到了98.42%、94.72%和97%。同时在Quick Draw涂鸦数据集上进行模型泛化能力的测试,本文所使用的模型在该数据集上展现出良好的提取效果。本研究的开展不仅扩充了微地图个性化点符号的研究数据集,还改进了通用地图符号的提取方法,为微地图制图注入了更多元化的元素,也为自媒体时代的地图制作提供了更为灵活和个性化的解决方案。
余懿韬 , 闫浩文 , 李精忠 , 王小龙 , 富璇 , 王卓 , 侯宇豪 . 面向微地图的手绘地图符号自动提取方法[J]. 地球信息科学学报, 2024 , 26(7) : 1646 -1658 . DOI: 10.12082/dqxxkx.2024.240110
The We-Map, a novel cartographic phenomenon emerging in the era of social media, is distinctively characterized by mass participation, personalization, and swift dissemination. However, existing research on We-Map falls short in addressing the intricate challenges posed by point symbol design, thereby hampering the fulfillment of the public's desire for personalized cartographic representations. To bridge this gap, this paper starts from the perspective of We-Map mapping production, taking common map symbols in hand-drawn maps as the research object, and constructs an open hand-drawn map dataset. To this end, we have constructed a comprehensive dataset encompassing a diverse array of hand-drawn map symbols, encompassing various types and styles. This dataset serves as a valuable resource for exploring and enhancing the automated extraction of common map symbols. Drawing inspiration from existing research, we have embarked on a journey to identify and evaluate the most suitable model for our task. Among the numerous models for object detection, the performance of the YOLOv5 series models is well-known, and therefore this article will not delve into it excessively. Specifically, through comparison, we ultimately chose the YOLOv5-X model, which boasts advanced capabilities in object detection and classification. By leveraging the YOLOv5-X model, we have achieved remarkable results in the automatic extraction of common map symbols from hand-drawn maps. Our experiments reveal that the model achieves high levels of accuracy, recall, and F1 score in identifying and extracting point categories from the hand-drawn map dataset. These scores stand testament to the model's effectiveness in capturing the intricate details and unique characteristics of hand-drawn map symbols. Moreover, to further validate the generalizability of our model, we have conducted additional experiments on the Quick Draw doodle dataset. The results obtained from these experiments confirm that our model performs equally well in extracting common map symbols from diverse and varying datasets. The significance of this study lies not only in enhancing the dataset available for personalized point symbol research in We-Map but also in advancing the techniques for extracting common map symbols. By introducing more diversified elements into We-Map cartography, we have opened up new avenues for more flexible and personalized mapmaking in the age of self-media. This study represents a significant step forward in the evolution of cartography in the rea of self-media, catering to the evolving needs and preferences of the modern audience. The finally extracted point symbols can provide a data foundation for downstream tasks related to We-Map.
表2 不同模型对数据集的识别精度Tab. 2 Recognition accuracy of different models on the dataset |
模型 | YOLOv5 | params/M | FLOPs@640/B | YOLOv8 | params/M | FLOPs@640/B |
---|---|---|---|---|---|---|
n | 28.0(300e) | 1.9 | 4.5 | 37.2(500e) | 3.2 | 8.7 |
s | 37.4(300e) | 7.2 | 16.5 | 44.9(500e) | 11.2 | 28.6 |
m | 45.4(300e) | 21.2 | 49.0 | 50.2(500e) | 25.9 | 78.9 |
l | 49.0(300e) | 46.5 | 109.1 | 52.9(500e) | 43.7 | 165.2 |
x | 50.7(300e) | 86.7 | 205.7 | 53.9(500e) | 68.2 | 257.8 |
表3 模型训练结果Tab. 3 Model training results |
YOLOv5版本 | 类型 | AP/% | 精确率/% | 召回率/% | F1/% |
---|---|---|---|---|---|
YOLOv5-S | point | 93.29 | 94.67 | 84.35 | 89 |
label | 80.16 | 92.57 | 56.91 | 70 | |
symbol | 58.67 | 86.48 | 42.58 | 57 | |
YOLOv5-L | point | 97.34 | 97.86 | 93.58 | 96 |
label | 90.88 | 95.41 | 75.89 | 85 | |
symbol | 74.01 | 90.98 | 59.80 | 72 | |
YOLOv5-M | point | 96.54 | 96.96 | 91.36 | 94 |
label | 86.74 | 93.95 | 70.82 | 81 | |
symbol | 68.72 | 89.15 | 51.10 | 67 | |
YOLOv5-X | point | 97.85 | 98.42 | 94.72 | 97 |
label | 91.59 | 96.04 | 78.61 | 86 | |
symbol | 77.29 | 92.05 | 63.55 | 75 |
表4 Quick Draw数据集子类识别结果Tab. 4 Quick Draw dataset subclass recognition results (%) |
类别 | 识别结果 | 类别 | 识别结果 | 类别 | 识别结果 | 类别 | 识别结果 |
---|---|---|---|---|---|---|---|
bat | 20.2 | cow | 59.0 | hot dog | 28.0 | radio | 44.0 |
bathtub | 59.8 | crayon | 14.6 | hurricane | 49.4 | sink | 52.0 |
cake | 85.0 | dragon | 22.0 | lantern | 36.4 | toaster | 81.8 |
camel | 26.6 | firetruck | 55.4 | monkey | 14.6 | zebra | 66.2 |
car | 26.4 | helmet | 3.2 | panda | 28.4 |
表5 与主流算法的对比Tab. 5 Comparison with mainstream algorithms (%) |
模型 | AP | 精确率 | 召回率 | F1 |
---|---|---|---|---|
Retinanet | 73.94 | 92.81 | 59.17 | 72 |
SSD | 68.77 | 92.04 | 36.03 | 52 |
YOLOv5-X | 97.85 | 98.42 | 94.72 | 97 |
[1] |
高俊. 换一个视角看地图[J]. 测绘通报, 2009(1):1-5.
[
|
[2] |
|
[3] |
王海鹰, 闫浩文, 田江鹏, 等. 后现代哲学视野下的微地图[J]. 武汉大学学报(信息科学版), 2022, 47(12):2026-2037.
[
|
[4] |
王家耀. 时空大数据时代的地图学[J]. 测绘学报, 2017, 46(10):1226-1237.
[
|
[5] |
闫浩文, 张黎明, 杜萍, 等. 自媒体时代的地图:微地图[J]. 测绘科学技术学报, 2016, 33(5):520-523.
[
|
[6] |
国家市场监督管理总局,国家标准化管理委员会. 公共地理信息通用地图符号: GB/T 24354—2023[S]. 北京: 中国标准出版社.
[ State Administration for Market Regulation of the People's Republic of China, Standardization Administration of the People's Republic of China. Common map symbols for the public geographical information: GB/T 24354—2023[S]. Beijing: Standards Press of China.]
|
[7] |
|
[8] |
宋伟轩, 吕陈, 徐旳. 城市社区微观空间意象研究——基于南京居民250份手绘草图的比较[J]. 地理研究, 2011, 30(4):709-722.
[
|
[9] |
申思, 薛露露, 刘瑜. 基于手绘草图的北京居民认知地图变形及因素分析[J]. 地理学报, 2008, 63(6):625-634.
[
|
[10] |
王娟, 李钢, 于悦, 等. 被拐儿童记忆地图模式及其影响因素——基于寻亲成功案例的实证[J]. 热带地理, 2022, 42(9):1559-1570.
[
|
[11] |
白娅兰, 闫浩文, 禄小敏, 等. 微地图符号的视觉变量及其应用[J]. 测绘科学, 2021, 46(7):182-188,204.
[
|
[12] |
闫晓婧, 闫浩文, 王小龙. 微地图符号的智能化匹配——“识图配符”[J]. 测绘科学, 2022, 47(10):205-212.
[
|
[13] |
何阳, 闫浩文, 王卓, 等. 面向微地图的地标提取方法及个性化寻路应用[J]. 地球信息科学学报, 2022, 24(5):827-836.
[
|
[14] |
|
[15] |
|
[16] |
富璇, 闫浩文, 王小龙, 等. 城市内涝场景下的微地图制作方法[J/OL]. 地球信息科学学报,1-14[2024-03-29].
[
|
[17] |
|
[18] |
Helmer,
|
[19] |
赵飞, 李兆正, 甘泉, 等. 结合SAM大模型和数学形态学的历史地图水系信息提取方法[J]. 测绘学报, 2024, 53(04):761-772.
[
|
[20] |
|
[21] |
|
[22] |
|
[23] |
|
[24] |
|
[25] |
|
[26] |
|
[27] |
|
[28] |
|
[29] |
|
[30] |
黄文骞. 数字地图符号的形状描述与识别[J]. 测绘学报, 1999, 3:233-238.
[
|
[31] |
俞连笙. 地图符号的哲学层面及其信息功能的开发[J]. 测绘学报, 1995, 4:259-266.
[
|
[32] |
|
[33] |
|
[34] |
|
[35] |
|
[36] |
|
[37] |
|
[38] |
|
[39] |
|
[40] |
|
[41] |
|
[42] |
|
/
〈 |
|
〉 |