基于Mask R-CNN的河道典型地物提取方法研究与应用  

Research and Application of Typical River Objects Extraction Method Based on Mask R-CNN

作　　者：王晟力 周子傲 吉红香[1,2,3] 郑泳 秦雁[1,2,3] 蔡季宏 WANG Shengli;ZHOU Ziao;JI Hongxiang;ZHENG Yong;QIN Yan;CAI Jihong(Guangdong Research Institute of Water Resources and Hydropower,Guangzhou 510635,China;Guangdong Key Laboratory of Hydrodynamic Research,Guangzhou 510635,China;State and Local Joint Engineering Laboratory of Estuarine Hydraulic Technology,Guangzhou 510635,China)

机构地区：[1]广东省水利水电科学研究院,广州510635 [2]广东省水动力学应用研究重点实验室,广州510635 [3]河口水利技术国家地方联合工程实验室,广州510635

出　　处：《广东水利水电》2024年第12期72-76,共5页Guangdong Water Resources and Hydropower

基　　金：2023年广东省科技创新战略专项资金科研项目(编号:gdsky2023-04)。

摘　　要：开展河湖遥感影像解译是提升河湖监管效能的迫切需要,也是智慧水利和数字孪生流域建设的重要内容,高时空分辨率遥感影像和深度学习算法可为快速识别监测河湖地物图斑、及时发现“四乱”问题和持续跟踪监测提供技术支撑。为摸查广东省主要河道水域岸线范围内典型地物底数,该文基于Mask R-CNN模型提出了一种河道水域岸线范围内桥梁和鱼塘自动识别方法,提取了广东省主要河道范围内两种典型地物,其中,桥梁787座,养殖鱼塘3773处,具有较高的准确性和实用性。结果表明,该方法对及时排查摸清涉河建设情况和河湖疑似问题图斑、常态化规范化推进河湖水域岸线动态监管具有一定应用前景。Remote sensing image interpretation of rivers and lakes is an urgent need to improve the efficiency of river and lake supervision,and it is also an important part of the construction of smart water conservancy and digital twin river basins.High spatiotemporal resolution remote sensing images and deep learning algorithms provide technical support for the rapid identification and monitoring of river and lake feature map spots,timely detection of the"four chaos"problems and continuous tracking and monitoring.In order to find out the number of typical objects within the shoreline of the main river waters in Guangdong Province,an automatic identification method of bridges and fish ponds within the shoreline of the river waters is proposed based on the Mask R-CNN model,and two typical ground features within the main river areas of Guangdong Province were extracted,including 787 bridges and 3773 fish ponds,which had high accuracy and practicability.The results show that this method has certain application prospects for timely investigation and mapping of river-related construction and suspected problems in rivers and lakes,and to promote the dynamic supervision of river and lake shorelines in a normalized and standardized manner.

关 键 词：河湖监管 Mask R-CNN 遥感解译 目标检测 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置] TV85[自动化与计算机技术—控制科学与工程]