兼顾要素正确分类及精准定位的栅格海图水深注记自动提取方法

doi:10.11947/j.AGCS.2023.20220258

摘要/Abstract

摘要： 针对当前水深注记自动提取实现困难、精度不高及效率过低等问题,将卷积神经网络(convolutional neural networks,CNN)模型应用于水深注记的自动识别,结合水深注记空间分布及几何特性对传统模式识别算法进行了改进,提出了一种兼顾要素正确分类及精准定位的栅格海图水深注记自动提取方法。通过对海图切片邻域的定量扩张,建立了顾及要素完整性的海图自适应切分模型,克服了CNN模型应用于大幅面海图要素识别的局限性;结合预测框角点位置的空间关系分析及评估,设计了面向空域冲突的要素唯一性判定原则,解决了邻域扩张引起的水深注记重复识别问题;在此基础上,进一步论证了水深注记主点位置的空间分布规律,建立了考虑要素几何分布特征的连通域分析改进模型,实现了水深注记的精准定位及数值提取。试验结果表明:①本文方法较好地实现了水深注记的自动提取,在CNN模型实现水深注记分类及粗定位过程中,具有较高的查全率和分类查准率。同时,最终水深注记数值提取结果正确率较高,且主点位置能满足水深注记提取的特殊要求。②通过多种CNN模型应用于本文自动提取模型中的对比试验,对比不同CNN模型在本文自动提取模型中的效能,分析并给出CNN模型采用的建议。同时,选取表现较好的CNN模型,作为本文模型采用的CNN模型与传统模式识别方法做对比,根据处理时间和查准、查全率结果,证明本文方法具有较高的识别准确率和效率。

关键词: 海图数字化, CNN, 深度学习, 水深注记, 自动提取

Abstract: Aiming at the problems of difficult implementation, low accuracy and low efficiency of automatic extraction of depth annotation, CNN model is applied to automatic recognition of depth annotations. Combined with the spatial distribution and geometric characteristics of depth annotation, the traditional pattern recognition algorithm is improved, a grid chart depth annotation automatic extraction method considering the correct classification and accurate positioning of elements is proposed. Through the quantitative expansion of the neighborhood of chart slices, an adaptive chart segmentation model considering the integrity of elements is established, which overcomes the limitations of CNN model applied to large format chart element recognition. Combined with the analysis and evaluation of the spatial relationship of the corner position of the prediction frame, the principle of determining the uniqueness of elements facing the airspace conflict is designed, and the problem of repeated recognition of depth annotation caused by neighborhood expansion is solved. On this basis, the spatial distribution law of the location of the main points of depth annotation is further demonstrated, and an improved connected domain analysis model considering the geometric distribution characteristics of elements is established, which realizes the accurate positioning and numerical extraction of depth annotation. The experimental results show that: ① This method can achieve the automatic extraction of depth annotation, and has high recall and precision in the process of classification and rough positioning of depth annotation based on CNN model. At the same time, the accuracy of the final depth annotation value extraction is high, and the position of the main point can meet the special requirements of depth annotation extraction; ② Through the comparative experiments of various CNN models applied in the automatic extraction model in this paper, the effectiveness of different CNN models in the automatic extraction model in this paper is compared, and the suggestions for the adoption of CNN models are analyzed and given. At the same time, the CNN model with good performance is selected as the CNN model used in this model and compared with the traditional pattern recognition method. According to the processing time and the accuracy recall results show that this method has high recognition accuracy and efficiency.

Key words: nautical chart, CNN, deep learning, sounding annotation, automatic extraction

中图分类号:

P227

马梦锴, 董箭, 唐露露, 彭认灿, 周寅飞, 王芳. 兼顾要素正确分类及精准定位的栅格海图水深注记自动提取方法[J]. 测绘学报, 2023, 52(6): 1022-1036.

MA Mengkai, DONG Jian, TANG Lulu, PENG Rencan, ZHOU Yinfei, WANG Fang. Automatic extraction method of depth annotation in grid chart considering correct classification and accurate positioning of elements[J]. Acta Geodaetica et Cartographica Sinica, 2023, 52(6): 1022-1036.

参考文献

[1] 宁津生, 陈俊勇, 李德仁, 等. 测绘学概论[M]. 武汉:武汉大学出版社, 2016. NING Jinsheng, CHEN Junyong, LI Deren, et al. Introduction to Geomatics[M]. Wuhan:Wuhan University Press, 2016.
[2] 董箭, 彭认灿, 张立华, 等. 数字水深优化建模及滚动球处理技术研究[M].北京: 测绘出版社, 2020. DONG Jian, PENG Rencan, ZHANG Lihua, et al. Research on Optimization Digital Depth Modeling and Rolling Ball Transform Processing[M]. Beijing:China Surveying and Mapping Press, 2020.
[3] 彭认灿, 于彩霞, 董箭, 等. 数字海图制图[M].北京: 测绘出版社, 2021. PENG Rencan, YU Caixia, DONG Jian, et al. Digital Chart Cartography[M].Beijing:China Surveying and Mapping Press, 2021.
[4] 孙立新, 黄明, 任美睿. 地图扫描影像中文字、符号注记的自动提取[J]. 测绘工程, 1997, 6(3): 26-30. SUN Lixin, HUANG Ming, REN Meirui. Automatic extracting of the text and symbol annotation in map scanning image[J]. Engineering of Surveying and Mapping, 1997, 6(3): 26-30.
[5] 徐战武, 张涛, 刘肖琳. 地形图数字注记的自动提取与识别[J]. 中文信息学报, 2001, 15(2): 51-56. XU Zhanwu, ZHANG Tao, LIU Xiaolin. Automatic extraction and recognition of numbers in topographic maps[J]. Journal of Chinese Information Processing, 2001, 15(2): 51-56.
[6] 陈睿, 张祖勋, 张剑清. 扫描地形图中数字高程注记的提取和识别[J]. 武汉大学学报(信息科学版), 2002, 27(2): 194-198. CHEN Rui, ZHANG Zuxun, ZHANG Jianqing. Identification of digital elevation annotation in scanned map[J]. Geomatics and Information Science of Wuhan University, 2002, 27(2): 194-198.
[7] 许鹏飞. 彩色地形图中地理要素提取与点状符号识别算法研究[D]. 西安: 西安电子科技大学, 2014. XU Pengfei. Research on the algorithm of geographical elements extraction and point symbol recognition in color topographic map[D]. Xi'an: Xidian University,2014.
[8] 沈意浪, 艾廷华, 赵荣. 一种彩色栅格地图注记识别方法[J]. 武汉大学学报(信息科学版), 2018, 43(1): 145-151. SHEN Yilang, AI Tinghua, ZHAO Rong. A method for color raster map annotation recognition[J]. Geomatics and Information Science of Wuhan University, 2018, 43(1): 145-151.
[9] 辛动军, 史迎春. 地图要素识别与提取研究现状[J]. 中州大学学报, 2009, 26(4): 115-118. XIN Dongjun, SHI Yingchun. Research status of map features' identification and extraction[J]. Journal of Zhongzhou University, 2009, 26(4): 115-118.
[10] GOODFELLOW I J, BULATOV Y, IBARZ J, et al. Multi-digit number recognition from street view imagery using deep convolutional neural networks[C]//Proceedings of 2014 International Conference on Learning Representations. Banff: ICLR,2014:1-13.
[11] 孟亦菲, 郑贵洲, 冀炜臻. 集成空间变换结构与深度残差网络的遥感影像场景分类方法[J/OL]. 地球科学:1-15[2021-12-16]. http://kns.cnki.net/kcms/detail/42.1874.P.20211125.1532.002.html. MENG Yifei, ZHENG Guizhou, JI Weizhen. Remote sensing image scene classification method integrating spatial transformation structure and sounding residual network[J/OL]. Earth Science: 1-15[2021-12-16]. http://kns.cnki.net/kcms/detail/42.1874.P.20211125.1532.002.html.
[12] 李典. 复杂场景下基于YOLOv3的人脸检测研究[D]. 广州:广东技术师范大学, 2021. LI Dian. Research on face detection in complex scenes based on YOLOv3[D]. Guangzhou: Guangdong Polytechnic Normal University, 2021.
[13] 王明吉, 刘博, 陈秋梦, 等. 基于YOLOv3的车牌定位识别系统[J]. 工业仪表与自动化装置, 2022(1): 97-100. WANG Mingji, LIU Bo, CHEN Qiumeng, et al. License plate location and recognition system based on YOLOv3[J]. Industrial Instrumentation & Automation, 2022(1): 97-100.
[14] 周飞燕, 金林鹏, 董军. 卷积神经网络研究综述[J]. 计算机学报, 2017, 40(6): 1229-1251. ZHOU Feiyan, JIN Linpeng, DONG Jun. Review of convolutional neural network[J]. Chinese Journal of Computers, 2017, 40(6): 1229-1251.
[15] GUO Qiang, LEI Jun, TU Dan, et al. Reading numbers in natural scene images with convolutional neural networks[C]//Proceedings of 2014 IEEE International Conference on Security, Pattern Analysis, and Cybernetics (SPAC).Wuhan: IEEE, 2014: 48-53.
[16] LI Huali, LIU Jun, ZHOU Xiran. Intelligent map reader: a framework for topographic map understanding with deep learning and gazetteer[J]. IEEE Access, 2018, 6: 25363-25376.
[17] 王家耀. 时空大数据时代的地图学[J]. 测绘学报, 2017, 46(10): 1226-1237. DOI: 10.11947/j.AGCS.2017.20170308. WANG Jiayao. Cartography in the age of spatio-temporal big data[J]. Acta Geodaetica et Cartographica Sinica, 2017, 46(10): 1226-1237. DOI: 10.11947/j.AGCS.2017.20170308.
[18] 温佩芝, 姚航, 沈嘉炜. 基于卷积神经网络的石刻书法字识别方法[J]. 计算机工程与设计, 2018, 39(3): 867-872. WEN Peizhi, YAO Hang, SHEN Jiawei. Recognition method of stone inscription font based on convolution neural network[J]. Computer Engineering and Design, 2018, 39(3): 867-872.
[19] 任福, 侯宛玥. 面向机器阅读的地图名称注记类别识别方法[J]. 武汉大学学报(信息科学版), 2020, 45(2): 273-280. REN Fu, HOU Wanyue. Identification method of map name annotation category for machine reading[J]. Geomatics and Information Science of Wuhan University, 2020, 45(2): 273-280.
[20] MADAKANNU A, SELVARAJ A. DIGI-Net: a deep convolutional neural network for multi-format digit recognition[J].Neural Computing and Applications, 2020, 32(15): 11373-11383.
[21] ASIF M, BIN AHMAD M, MUSHTAQ S, et al. Long multi-digit number recognition from images empowered by deep convolutional neural networks[J]. The Computer Journal, 2022, 65(10): 2815-2827.
[22] 艾廷华.深度学习赋能地图制图的若干思考[J]. 测绘学报, 2021, 50(9): 1170-1182.DOI:10.11947/j.AGCS.2021.20210091. AI Tinghua. Some thoughts on deep learning empowerment map drawing[J]. Acta Geodaetica et Cartographica Sinica, 2021, 50(9): 1170-1182. DOI:10.11947/j.AGCS.2021.20210091.
[23] 周熙然,李德仁,薛勇,等.地图图像智能识别与理解:特征、方法与展望[J/OL].武汉大学学报(信息科学版),2022,47(05):641-650. DOI:10.13203/j.whugis20210300. ZHOU Xiran, LI Deren, XUE Yong,et al. GeoAI framework of intelligent recognition for ubiquitous map imagery: current state and prospect[J/OL]. Geomatics and Information Science of Wuhan University,2022,47(05):641-650. DOI:10.13203/j.whugis20210300.
[24] 周熙然, 李德仁, 薛勇, 等. 地图图像智能识别与理解: 特征、方法与展望[J]. 武汉大学学报(信息科学版), 2022, 47(5): 641-650. ZHOU Xiran, LI Deren, XUE Yong, et al. Intelligent map image recognition and understanding: representative features, methodology and prospects[J]. Geomatics and Information Science of Wuhan University, 2022, 47(5): 641-650.
[25] REDMON J, FARHADI A. YOLOv3: an incremental improvement[C]//Proceedings of 2018 Conference on Computer Vision and Pattern Recognition (CVPR). Salt Lake City: IEEE,2018: 85-89.
[26] 蒲玮. 基于YOLOv3的遥感图像目标检测算法研究[D]. 天津:河北工业大学, 2020. PU Wei. Research on target detection algorithm of remote sensing image based on YOLOv3[D]. Tianjin: Hebei University of Technology, 2020.
[27] 赵永强, 饶元, 董世鹏, 等. 深度学习目标检测方法综述[J]. 中国图象图形学报, 2020, 25(4): 629-654. ZHAO Yongqiang, RAO Yuan, DONG Shipeng, et al. Survey on deep learning object detection[J]. Journal of Image and Graphics, 2020, 25(4): 629-654.
[28] 张俊, 杨光, 胡东升, 等. 基于注意力机制与YOLOv3的电表自动读数算法[J]. 电力信息与通信技术, 2021, 19(12): 82-87. ZHANG Jun, YANG Guang, HU Dongsheng, et al. An automatic reading algorithm of power meter based on YOLOv3 and attention mechanism[J]. Electric Power Information and Communication Technology, 2021, 19(12): 82-87.
[29] 蔡兴泉, 阮瓒茜, 孙海燕. 基于YOLOv3和MobileNetv2的银行卡号识别方法[J]. 计算机辅助设计与图形学学报, 2022, 34(1): 142-151. CAI Xingquan, RUAN Zanxi, SUN Haiyan. Bank card number identification method based on YOLOv3 and MobileNetv2[J]. Journal of Computer-Aided Design & Computer Graphics, 2022, 34(1): 142-151.
[30] 贺智龙. 基于YOLOv3的车牌识别研究[D]. 济南:齐鲁工业大学, 2021. HE Zhilong. Research on license plate recognition based on YOLOv3[D]. Jinan: Qilu University of Technology, 2021.
[31] 张婷婷, 马明栋, 王得玉. OCR文字识别技术的研究[J]. 计算机技术与发展, 2020, 30(4): 85-88. ZHANG Tingting, MA Mingdong, WANG Deyu. Research on OCR technology[J]. Computer Technology and Development, 2020, 30(4): 85-88.
[32] United Kingdom Hydrographic Office.Symbols and Abbreviations used on Admiralty paper Charts Ed4: UKHO-5011[S]. taunton:[s.n.], 2008.
[33] 嵇新浩. 基于连通域的文本定位方法研究[D]. 杭州:浙江工业大学, 2007. JI Xinhao. Connected component based approach for text localization in images[D]. Hangzhou: Zhejiang University of Technology, 2007.
[34] 张羽. 目标检测中的交并比损失函数研究[D]. 合肥:安徽大学, 2021. ZHANG Yu. Research on IOU loss function in target detection[D]. Hefei: Anhui University, 2021.
[35] HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 770-778.
[36] 国家质量技术监督局.中国航海图书编绘规范:GB 12320—1998 [S].北京: 中国标准出版社, 1999. The State Bureau of Quality and Technical Supervision.Specifications for Chinese nautical charts: GB 12320—1998 [S].Beijing: China Standard Press, 1999.
[37] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[38] LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[C]//Proceedings of 2016 Computer Vision-ECCV. Cham: Springer International Publishing, 2016: 21-37.
[39] ZHU Xingkui, LYU Shuchang, WANG Xu, et al. TPH-YOLOv5: improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). Montreal: IEEE, 2021: 2778-2788.
[40] LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu: IEEE, 2017: 936-944.
[41] 朱强强. 面向复杂场景的小目标检测识别深度网络模型研究[D]. 南昌:江西科技师范大学, 2021. ZHU Qiangqiang. Research on the deep network model of small target detection and recognition for complex scene[D]. Nanchang: Jiangxi Normal University of Science and Technology, 2021.