基于深度学习的堰塞坝表层颗粒物质识别研究  

Research on surface particle matter identification of barrier dam based on deep learning

作　　者：付兵杰 李书 丁凡桠[1,2] 栾约生 Fu Bingjie;Li Shu;Ding Fanya;Luan Yuesheng(Changjiang Survey,Planning,Design and Research Co.,Ltd.,Wuhan 430010,China;Changjiang Survey Institute of MWR,Wuhan 430011,China)

机构地区：[1]长江勘测规划设计研究有限责任公司,武汉430010 [2]水利部长江勘测技术研究所,武汉430011

出　　处：《工程勘察》2025年第2期79-84,共6页Geotechnical Investigation & Surveying

基　　金：长江勘测规划设计研究有限责任公司2021年度自主创新项目“基于深度学习的堰塞坝表层颗粒物质智能识别和分析关键技术研究与应用”(CX2021Z72).

摘　　要：在堰塞坝形成初期,交通闭塞,人员无法快速进入的情况下,利用无人机影像对堰塞坝表层颗粒物质进行智能识别,对于堰塞坝表层结构分析和危险性评价具有重要意义。堰塞坝表层颗粒物质一般分布散乱、粒径不均匀、背景复杂,利用人工解译或传统图像识别技术存在自动化程度低、识别率差等问题。针对上述不足,提出一种基于深度学习的堰塞坝表层颗粒物质识别方法,采用“大颗粒逐个识别→密集小颗粒逐分割块识别”的思路,利用Mask R-CNN、SLIC、AlexNet等方法联合构建自动识别模型,并进行精度评价和推广应用实验。结果表明,该方法可实现对堰塞坝表层颗粒物质的智能识别和粒径分析等功能,识别精度大于90%,具有一定的迁移能力,能够为堰塞坝应急抢险提供较好的数据基础和决策依据。In the early stage of barrier dam formation,when traffic is blocked and people cannot enter quickly,intelligent identification of particle matter on the surface of the barrier dam using UAV images is of great significance for the surface structure analysis and risk assessment of the barrier dam.The surface particle matter of barrier dam are generally scattered,with uneven particle size and complex background.There are some problems such as low degree of automation and poor recognition rate when using manual interpretation or traditional image recognition technology.In order to solve the above problems,a method based on deep learning is proposed for the identification of the particle matter.With the idea of"one by one identification of large particles and one by one identification of dense small particles block",the automatic recognition model is constructed by using Mask R-CNN,SLIC,AlexNet,etc.,and the accuracy evaluation and application experiments are carried out.The results show that the proposed method can realize the functions of intelligent identification and particle size analysis of the surface particle matter of the barrier dam,the identification accuracy is greater than 90%,and it has certain migration ability.It can provide a good data basis and decision-making support for the emergency rescue of the barrier dam.

关 键 词：堰塞坝 表层颗粒物质 深度学习 图像识别 

分 类 号：P237[天文地球—摄影测量与遥感]