• ▶ 2008-2024年被中国情报信息研究所评价中心评为“中国科技核心期刊”
  • ▶ 2019-2024年连续三届被中国科学院文献情报中心中国科学引文数据库CSCD(核心库)收录
  • ▶ 2021、2023年入编北京大学图书馆《中文核心期刊要目总览》
  • ▶ 2020-2024连续四年入选《科技期刊世界影响力指数(WJCI)报告》
刘东海, 马子茹, 黄 斌, 刘雅雯, 王志岗.基于无人机巡检与深度学习的河道整治施工进度图像识别水资源与水工程学报[J].,2024,35(4):92-100
基于无人机巡检与深度学习的河道整治施工进度图像识别
Image recognition of river regulation construction progress based on UAV inspection and deep learning
  
DOI:10.11705/j.issn.1672-643X.2024.04.11
中文关键词:  河道整治工程  施工进度  无人机巡检  图像识别  目标检测  特征点匹配
英文关键词:river regulation project  construction progress  unmanned air vehicle (UAV) inspection  image recognition  object detection  feature point matching
基金项目:中国长江三峡集团有限公司企业科研项目(202103551)
作者单位
刘东海1, 马子茹1, 黄 斌2, 刘雅雯2, 王志岗2 (1.天津大学 水利工程智能建设与运维全国重点实验室 天津 300350 2.长江三峡技术经济发展有限公司 北京 101100) 
摘要点击次数: 1110
全文下载次数: 740
中文摘要:
      长线性小流域治理工程中的河道衬砌、生态护岸等距离长、范围广、布置分散,且工区交通不便,人工巡检费时费力,难以及时掌握工程的整体施工进度形象面貌。提出了基于无人机巡检与深度学习的河道整治施工进度智能图像识别方法,通过定位施工节点(施工区域起点和终点)的位置计算施工进度。首先,建立了施工区域目标检测模型,针对无人机航拍影像进行河道衬砌护岸施工区域的识别以及施工节点的定位;然后,利用尺度不变特征变换(scale-invariant feature transform, SIFT)算法对不同视频帧中的施工节点进行匹配,并基于单目视觉的运动视差法,计算施工节点的实际工区坐标;最后,计算当前衬砌护岸施工进度,并分析进度偏差。结果表明:该方法得到的施工节点定位平均误差为1.026 m,平均相对误差为0.74%,该方法能够较为准确地从航拍图像中识别得到当前衬砌护岸的施工进程,从而实现长线性工区快速巡检,及时掌控现场施工进度,提高工程管理的智能化水平。
英文摘要:
      Long-line river regulation projects such as bank lining and revetment projects are far stretched and scattered with a wide range, resulting in inconvenient transportation for inspection. Besides, conventional manual inspection is time-consuming and labor-intensive, which is difficult to grasp the construction progress of the whole project in practice. In view of this, this paper proposes an intelligent construction progress monitoring method of river regulation projects based on unmanned air vehicle (UAV) inspection and deep learning, which can calculate the construction progress by locating construction nodes (i.e., starting point and ending point of the construction area). Firstly, the object detection model of construction area is established to recognize the construction area of lining and revetment and locate the construction nodes based on UAV aerial photography. Then, SIFT (scale-invariant feature transform) algorithm is used to match the construction nodes in different video frames, and a motion parallax method based on monocular vision is introduced to locate the actual work area coordinates of construction nodes. Finally, the current lining construction progress is calculated and the progress deviation is analyzed. The results show that the average error of the construction progress is 1.026 m, and the average relative error is 0.74 %, indicating that the proposed method can recognize the construction progress of lining and revetment accurately based on the UAV images, thereby achieving full coverage and rapid inspection of long-line projects, timely control of on-site construction progress, and improvement of the intelligent level of engineering management.
查看全文  查看/发表评论  下载PDF阅读器
关闭