基于改进Unet网络的河流表面流速识别方法

周继威; 安国成; 王根一

doi:10.12454/j.jsuese.202400869

您当前的位置：

首页 >

文章列表页 >

基于改进Unet网络的河流表面流速识别方法

更新时间：2025-05-20

- 基于改进Unet网络的河流表面流速识别方法
- An Improved U-Net-Based Framework for Estimating River Surface Flow Velocity
- 工程科学与技术 2025年页码：1-12
- 作者机构：
  
  1.龙源（北京）新能源工程技术有限公司，北京 100034
  2.上海华讯网络系统有限公司服务运作部，成都 6100741
- 作者简介：
  
  周继威（1981—），男，正高工程师，硕士. 研究方向：新能源数字化；人工智能. E-mail：12031963@ceic.com
  安国成，E-mail：anguocheng@eccom.com.cn
- 基金信息：
  
  国家重点研发计划资助项目(2023YFC3006700)
- DOI：10.12454/j.jsuese.202400869
  中图分类号： TP183
- 收稿日期：2024-10-28，
  
  修回日期：2025-04-18，
  
  网络出版日期：2025-05-19，
- 稿件说明：
移动端阅览
.An Improved U-Net-Based Framework for Estimating River Surface Flow Velocity[J].Advanced Engineering Sciences,XXXX,XX(XX):1‒12.周继威,安国成,王根一.基于改进Unet网络的河流表面流速识别方法[J].工程科学与技术,

An Improved U-Net-Based Framework for Estimating River Surface Flow Velocity[J/OL]. Advanced engineering sciences, 2025, 1-12.
.An Improved U-Net-Based Framework for Estimating River Surface Flow Velocity[J].Advanced Engineering Sciences,XXXX,XX(XX):1‒12.周继威,安国成,王根一.基于改进Unet网络的河流表面流速识别方法[J].工程科学与技术, DOI：10.12454/j.jsuese.202400869.

An Improved U-Net-Based Framework for Estimating River Surface Flow Velocity[J/OL]. Advanced engineering sciences, 2025, 1-12. DOI： 10.12454/j.jsuese.202400869.

摘要

河流表面流速的精准识别是水力发电效能优化的关键基础，针对复杂河流环境的数据采集差异性及少样本训练导致该领域深度学习模型泛化性不足的问题，本研究提出一种基于改进Unet网络的河流表面流速识别方法，旨在通过优化标签构建、训练策略及网络架构提升流速识别的精度与鲁棒性。本研究聚焦3个核心问题：1）基于钟型与阶梯式分布特性生成高精度灰度标签，将河道中心至边缘的流速梯度映射为灰度值，结合多摄像头多分辨率（2 560×1 440、1 920×1 080、1 280×720）数据采集构建跨场景数据集（河海实验室、Belgium）；2）设计双输出分支Unet网络，通过编码器

-解码器对称结构提取多尺度特征，在解码阶段引入融合层与解构层实现时间维度信息融合与通道匹配，同时输出2维流速分布图和1维像素位移值，结合双重监督损失函数优化模型；3）采用时空训练法，以连续32帧视频序列为输入，结合多尺度随机裁剪（160~640像素）及沿河道中心纵向掩码增强（掩码B，

=0.1）提升动态特征捕捉能力。实验表明，本方法在河海数据集低、中、高流速段（0~1、1~2、2~3 m/s）的RMSE分别为0.069、0.103、0.238，较Fast光流法（0.794）、CBEGAN（0.463）及Two-stage-method（0.297）降低50%~80%；Belgium数据集

MSE

为0.059（

=0.047），流速分布匹配精度（

AVD

）与准确率（

）达0.892和0.942；单帧推理时间22 ms（较CBEGAN减少67%），为水力发电效能优化、洪水预警等场景提供了可靠数据支持。

Abstract

Objective

Accurate estimation of river surface flow velocity is critical for optimizing hydropower generation efficiency and enhancing flood warning systems. Existing deep learning models face challenges in generalization due to limited training samples and data heterogeneity across complex river environments. This study aimed to address these limitations by developing a robust framework that integrates improved data labeling strategies

optimized training methods

and a novel neural network architecture to achieve high-precision velocity estimation.

Methods

Methods A dual-output branch structure was designed within an improved U-Net architecture to simultaneously supervise velocity distribution and pixel displacement values

enhancing feature representation and model robustness. The framework incorporated a spatiotemporal training strategy

where continuous video sequences were utilized to capture temporal dynamics. Data preprocessing included multi-resolution image acquisition (2560×1 440

1 920×1 080

1 280×720) from simulated river channels and natural environments. Two label generation methods—bell-shaped and stepped distributions—were proposed to encode velocity information into grayscale maps

balancing precision and computational efficiency. Multi-scale random cropping (160~640 pixels) and a customized masking strategy were applied to augment training data

focusing on critical river channel regions. Model performance was evaluated using RMSE

MAVD (matching accuracy of velocity distribution)

and FRA (Flow Rate Accuracy) metrics across the Hehai

Belgium

and real-world CDJJB datasets.

Results and Discussions

The proposed framework demonstrated superior performance across diverse datasets

achieving an average

MSE

of 0.137 on the Hehai dataset (subranges: H(0-1)=0.069

H(1-2)=0.103

H(2-3)=0.238) and M

=0.109

outperforming Fast optical flow (RMSE=0.272)

CBEGAN ( RMSE=0.217)

and the Two-stage method ( RMSE=0.183) by 50%~80%. On the Belgium dataset

RMSE and MAE were reduced to 0.059 and 0.047

respectively

while cross-dataset validation on real-world CDJJB data confirmed robustness (RMSE=0.086

MAE=0.073) under natural turbulence and lighting variations. These results validated the framework’s ability to address data heterogeneity and limited training samples through integrated spatiotemporal training and adaptive augmentation.Multi-scale random cropping (160~640 pixels) significantly enhanced velocity recognition accuracy. Larger crops (640 pixels) preserved spatial context

achieving MAVD=0.862 and FRA=0.927

whereas smaller crops (160 pixels) limited velocity detection to 1.5 m/s

inadequate for high-flow scenarios (

3 m/s). Intermediate scales (320 pixels) balanced computational efficiency and accuracy (MAVD=0.816)

with training loss curves revealing accelerated convergence for 640-pixel inputs

reducing loss saturation by 40% compared to 160-pixel crops. Customized masking strategies further improved precision by focusing on critical river regions. Mask B

aligned with high-velocity zones

achieved MAVD=0.892 and FRA=0.942

surpassing random masking (MAVD=0.881) by 4% and reducing prediction uncertainty by 23% in flows

2 m/s. In contrast

random masking (k=3

s=0.3) degraded accuracy (MAVD=0.722) due to excessive occlusion of hydrodynamic features

underscoring the nec

essity of domain-specific augmentation.Temporal sequence optimization revealed N=32 as the optimal video length

balancing spatiotemporal feature extraction (MAVD=0.816

FRA=0.894). Longer sequences (N=48/64) introduced redundancy (R=0.35~0.42)

degrading MAVD by 8%–12% and increasing inference time by 72% (N=64: 38 ms vs. N=32: 22 ms). Information entropy analysis confirmed redundant features (R

0.3) in extended sequences increased computational complexity without improving accuracy. Comparative analysis with state-of-the-art methods highlighted the framework’s advantages. For high-velocity flows (

2 m/s)

the dual-output architecture reduced RMSE to 0.238

a 70% improvement over traditional optical flow (OTV: 0.794). Real-time inference at 22 ms/frame surpassed CBEGAN (67 ms) and the Two-stage method (41 ms). On the Belgium dataset

MAE=0.047 in low-flow conditions (

0.8 m/s) outperformed OTV (0.074) and CBEGAN (0.095). Cross-dataset validation under natural turbulence (CDJJB) maintained MAE=0.073

demonstrating adaptability to environmental heterogeneity.The integration of adaptive label generation (bell-shaped/stepped distributions) and spatiotemporal training addressed data scarcity by leveraging sequential video dynamics. Mask B’s longitudinal alignment with high-velocity zones ensured focused learning on hydrodynamic features

while multi-scale cropping enhanced generalization through spatial context retention.. These results validate the framework’s potential for optimizing energy output and flood management

while emphasizing the necessity of expanded datasets to address environmental variability.

Conclusions

The improved U-Net framework effectively addresses challenges in river surface velocity estimation by integrating spatiotemporal training

adaptive labeling

and data augmentation. The dual-output structure ensures accurate velocity mapping and displacement prediction

while multi-scale cropping and targeted masking enhance generalization. Experimental results validate the method’s efficiency and accuracy across diverse datasets

with significant implications for hydropower optimization and flood management. Limitations in handling low-light conditions highlight the need for nighttime dataset expansion in future work.

关键词

Keywords

references

Xu L , Zhang Z , Yan X X , et al . Advances of non-contact instruments and techniques for open-channel flow measurements [J ] . Water Resource Information , 2013 , 3 : 37 – 44 .

Chen huayong , Hu Kuaiheng , Cui Peng , et al . Experimental Investigation on the Inner Distribution of Debris Flow Velocity [J ] . Advanced Engineering Sciences , 2015 , 47 ( 6 ): 77 - 83 . DOI: 10.15961/j.jsuese.2015.06.011 http://dx.doi.org/10.15961/j.jsuese.2015.06.011

陈华勇 , 胡凯衡 , 崔鹏 , 周公旦 . 泥石流内部流速分布规律实验研究 [J ] . 工程科学与技术 , 2015 , 47 ( 6 ): 77 - 83 . DOI: 10.15961/j.jsuese.2015.06.011 http://dx.doi.org/10.15961/j.jsuese.2015.06.011

Zhao H , Chen H , et al . Application of flow measurement method based on space-time image recognition of river surface [J ] . Journal of Water Resources Research , 2020 , 9 ( 1 ): 1 – 11 . DOI: 10.12677/JWRR.2020.91001 http://dx.doi.org/10.12677/JWRR.2020.91001

Jia Tianlong . Research on multi-objective optimal operation of reservoir power generation and navigation [D ] . Huazhong University of Science and Technology , 2020 .

贾天龙 . 水库发电-通航多目标优化调度研究 [D ] . 华中科技大学 , 2020 .

Fujita I , Muste M , & KRUGER A . Large-scale particle image velocimetry for flow analysis in hydraulic engineering applications [J ] . Journal of Hydraulic Research , 1998 , 36 ( 3 ), 397 – 414 . DOI: 10.1080/00221689809498605 http://dx.doi.org/10.1080/00221689809498605

Fujita I , Watababe H. , & Tsubaki R . Development of a non-intrusive and efficient flow monitoring technique: The space-time image velocimetry (STIV) [J ] . International Journal of River Basin Management , 2007 , 5 ( 2 ), 105 – 114 .

Le COZ J. , Jodeau M , Hauel A , Marchamd B , Leboursicaud BOURSICAUD R . Image-based velocity and discharge measurements in field and laboratory river engineering studies using the free FUDAA-LSPIV software [C ] // Proceedings of the international conference on fluvial hydraulics , river flow , 2014 , 3 : 05 . DOI: 10.1201/b17133-262 http://dx.doi.org/10.1201/b17133-262

Patalano A , Garcia C M , Ridriguez A . Rectification of image velocity results (river): A simple and user-friendly toolbox for large scale water surface particle image velocimetry (PIV) and particle tracking velocimetry (PTV) [J ] . Computers and Geosciences , 2017 , 109 , 323 – 330 . DOI: 10.1016/j.cageo.2017.09.001 http://dx.doi.org/10.1016/j.cageo.2017.09.001

Thielicke W , Stamhuis E . PIVlab–towards user-friendly, affordable and accurate digital particle image velocimetry in MATLAB [J ] . Journal of Open Research Software , 2014 , 2 ( 1 ). DOI: 10.5334/jors.bl http://dx.doi.org/10.5334/jors.bl

Tauro F , Tosi F , Mattoccia S , et al . Optical tracking velocimetry (OTV): Leveraging optical flow and trajectory-based filtering for surface streamflow observations [J ] . Remote Sensing , 2018 , 10 ( 12 ), 2010 . DOI: 10.3390/rs10122010 http://dx.doi.org/10.3390/rs10122010

Zhang Z , Zhou Y , Li X R , Chen H , Liu L H . Development and application of image flow measurement system [J ] . Water Conservancy Informatization 2018 ( 03 ), 7 - 13 .

Wang Wanliang , Qiu Hong , Zheng Jianwei . Estimation method of river surface velocity based on compressed sensing image analysis [J ] . Journal of Hydroelectric Engineering , 2018 , 37 ( 5 ), 69 - 79 .

王万良 , 邱虹 , 郑建炜 . 基于压缩感知图像分析的河流表面流速估计方法 [J ] . 水利水电工程杂志 , 2018 , 37 ( 5 ), 69 - 79 .

Chen W , Mied R. P . Optical flow estimation for motion-compensated compression [J ] . Image and Vision Computing , 2013 , 31 ( 3 ), 275- - 289 . DOI: 10.1016/j.imavis.2012.10.004 http://dx.doi.org/10.1016/j.imavis.2012.10.004

Jyoti J , S. Medeiros H , Sebo S , and Mcdonald W . " Remote sensing of river velocity using drone video and optical flow algorithm," in Watershed Management [J ] . American Society of Civil Engineers Reston , VA, 2020 , 97 – 204 .

Jyoti J S , Medeiros H , Sebo S , and Mcdonald W . River velocity measurements using optical flow algorithm and unoccupied aerial vehicles: A case study [J ] . Flow Measurement and Instrumentation , 2023 , 91 , 102314 . DOI: 10.1016/j.flowmeasinst.2023.102314 http://dx.doi.org/10.1016/j.flowmeasinst.2023.102314

Yao Qiang . Research on a method for detecting river surface velocity based on optical flow estimation [D ] . Zhejiang University , 2022 .

姚强 . 基于光流估计的河道表面流速检测方法研究 [D ] . 浙江大学 , 2022 .

Wang Wanliang , Yang Shenglan , Zhao Yanwei , et al . Estimation of river surface velocity based on conditional boundary balance generation adversarial network [J ] . Journal of Zhejiang University: Engineering Edition , 2019 , 53 ( 11 ): 11 . DOI: 10.3785/j.issn.1008-973X.2019.11.009 http://dx.doi.org/10.3785/j.issn.1008-973X.2019.11.009

王万良 , 杨胜兰 , 赵燕伟 , 等 . 基于条件边界平衡生成对抗网络的河流表面流速估测 [J ] . 浙江大学学报:工学版 , 2019 , 53 ( 11 ): 11 . DOI: 10.3785/j.issn.1008-973X.2019.11.009 http://dx.doi.org/10.3785/j.issn.1008-973X.2019.11.009

Wang X L , Ma Q , Wang G Y , et al . Video velocity measurement: A two-stage flow velocity prediction method based on deep learning [J ] . Hydrology Research , 2024 , 55 ( 5 ): 519 – 536 . DOI: 10.2166/nh.2024.012 http://dx.doi.org/10.2166/nh.2024.012

Farneback G . Two-frame motion estimation based on polynomial expansion [J ] . Lecture Notes in Computer Science , 2003 , 2749 , 363 - 370 . DOI: 10.1007/978-3-540-45179-2_30 http://dx.doi.org/10.1007/978-3-540-45179-2_30

Liu J , Zu J , Zhang Y , et al . Optical flow estimation method under the condition of ilumination change [J ] . Journal of Ilmage and Graphics , 2014 , 19 ( 10 ), 1475 - 1480 .

WU Y X , ZHANG J B , CAO Y Q , et al . River Surface Velocimetry Based on Virtual River Dataset and Modulated-RAFT . IEEE Access , DOI: 10.1109/ACCESS.2023.3267635 http://dx.doi.org/10.1109/ACCESS.2023.3267635 .

Ronneberger O , Fischer P , Brox T . U-Net: Convolutional networks for biomedical image segmentation [C ] // Proceedings of the 18th International Conference on Medical Image Computing and Computer-assisted Intervention . Munich : Springer , 2015 , 234 – 241 . DOI: 10.1007/978-3-319-24574-4_28 http://dx.doi.org/10.1007/978-3-319-24574-4_28

Li X , Erpicum S , Mignot E , Archambeau P , et al ( 2021 ). Scientific data: Laboratory modelling of urban flooding (v1) [Data set] . Zenodo . https://doi.org/10.5281/zenodo.5254164.DOI:10.5281/zenodo.5254164 https://doi.org/10.5281/zenodo.5254164.DOI:10.5281/zenodo.5254164

Yang Dan , Shao Guangjun , Hu Weifei , et al . A review of image-based river surface velocimetry [J ] . Journal of Zhejiang University , Engineering Edition, 2021 , 55 ( 9 ): 1752 - 1763 . DOI: 10.3785/j.issn.1008-973X.2021.09.017 http://dx.doi.org/10.3785/j.issn.1008-973X.2021.09.017

杨聃 , 邵广俊 , 胡伟飞 , 等 . 基于图像的河流表面测速研究综述 [J ] . 浙江大学学报 ,工学版, 2021 , 55 ( 9 ): 1752 - 1763 . DOI: 10.3785/j.issn.1008-973X.2021.09.017 http://dx.doi.org/10.3785/j.issn.1008-973X.2021.09.017

Rosyen E . Machine learning for high-speed corner detection [C ] . European Conference on Computer Vision Springer-Verlag , 2006 . DOI: 10.1007/11744047_19 http://dx.doi.org/10.1007/11744047_19

胡鑫雯 ; 王浩楠 ; 郑睿一种基于特征对象和透视变换的图像距离信息提取方 : CN 201911388812.7 [P ] . 2019-12-30 .

Tan M , Pang R , Le Q V . EfficientDet: Scalable and Efficient Object Detection [C ] // IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . IEEE , 2020 : 10781 – 10790 . DOI: 10.1109/CVPR42600.2020.01080 http://dx.doi.org/10.1109/CVPR42600.2020.01080

Teed Z , Deng J . Raft: Recurrent all-pairs field transforms for optical flow [C ] // Proceedings of the European Conference on Computer Vision , 2021 : 402 – 419 . DOI: 10.1007/978-3-030-58536-0_24 http://dx.doi.org/10.1007/978-3-030-58536-0_24

Liu xiaxuande , Shen Danfeng , Zhang Xuxiang , et al . The improved LK optical flow method is used to track moving ball targets in complex environments [J ] . Computer system applications , 2019 , 28 ( 7 ): 221 - 227 . DOI: 10.15888/j.cnki.csa.006973 http://dx.doi.org/10.15888/j.cnki.csa.006973

刘夏轩德 , 沈丹峰 , 张旭祥 , 张国英 . 改进LK光流法在复杂环境中对移动小球目标追踪 . 计算机系统应用 , 2019 , 28 ( 7 ): 221 - 227 . DOI: 10.15888/j.cnki.csa.006973 http://dx.doi.org/10.15888/j.cnki.csa.006973

浏览量

下载量

CNKI被引量

文章被引用时，请邮件提醒。

提交

工具集

关联资源

暂无数据