1 State Key Laboratory for Tunnel Engineering, School of Civil Engineering, Shandong University, Jinan 250061, China
2 China Power Construction Group Guiyang Survey Design and Research Institute Co., Ltd., Guiyang 550000, China
Show less
文章历史+
Received
Published
Revised
2025-06-17
2026-06-08
2025-11-28
Online Date
2026-08-07
PDF (9930K)
摘要
堤坝隐患识别在水利工程安全监测中具有重要意义。然而,由于物探数据来源多样、特征差异显著以及目标区域边界模糊,现有方法在多源物探数据融合与精细识别方面仍面临挑战。针对上述问题,提出了一种基于改进U型神经网络(U-Net neural network,NetU-Net)的多源协同识别方法,设计了独立特征提取模块与噪声特征辅助机制,用于增强瞬变电磁、探地雷达与水下声呐数据在特征层的表达一致性与互补性。通过增强跳跃连接结构、引入加权交叉熵损失函数及余弦退火学习率调度策略,提升了模型对小样本类别、高噪声背景和隐患边界的辨识能力。在江西九江市立新圩堤坝实测数据上的应用表明,该方法在典型隐患区域的识别准确率达到87%以上,在均交并比(mean intersection over union,mIoU)等指标上相较传统方法取得明显提升。研究结果验证了所提方法在多源物探数据融合处理与复杂地质条件下的稳定性与适应性,可为堤坝隐患识别系统的智能化与工程化提供技术支撑。
Abstract
Dam hazard identification plays a critical role in water conservancy safety monitoring. However, existing methods encounter significant challenges in multi-source geophysical data fusion and precise identification due to diverse data sources, substantial feature variations, and ambiguous target boundaries. To address these issues, an improved U-Net neural network (U-Net)-based multi-source collaborative identification framework was proposed. Independent feature extraction modules coupled with noise-assisted mechanisms were designed to enhance feature consistency and complementarity across transient electromagnetic, ground-penetrating radar, and underwater sonar data. The model’s capability in identifying minority classes, high-noise backgrounds, and hazard boundaries was strengthened through enhanced skip connections, weighted cross-entropy loss function, and cosine annealing learning rate scheduling. Field application at the Lixinxu embankment in Jiujiang City, Jiangxi Province demonstrates that the proposed method achieves over 87% recognition accuracy in typical hazard zones, with significant improvements in mean intersection over union (mIoU) and other metrics compared to conventional approaches. The results validate the method’s stability and adaptability in processing multi-source geophysical data under complex geological conditions, providing reliable technical support for intelligent dam hazard identification systems.
ZHANG Duo, WANG Bo, DENG Zhao-yang, ZHANG Shuo, XIAO Yue, NIE Li-chao.
Intelligent Comprehensive Detection of Dam Defects Based on Improved U-Net Neural Network[J].
Science Technology and Engineering, 2026, 26(16): 6749-6760 DOI:10.12404/j.issn.1671-1815.2504517
针对探地雷达(ground penetrating radar, GPR)、瞬变电磁法(transient electromagnetic method, TEM)[26]和多波束声呐数据的异构性特点,构建基于专属CNN的独立特征提取模块。针对多源物探数据融合中信息冲突与特征丢失的问题,采用模块化设计,逐层提取各数据源的关键特征,实现特征解耦与精细表达,避免直接融合带来的冗余和冲突。基于非线性激活机制,模块引入线性整流函数(rectified linear unit, ReLU)激活函数,通过从线性映射到非线性空间的转换,增强了模型对管涌水通道、富水区和坝体松散区等复杂隐患区域的特征表达能力。此外,为解决计算复杂度高和特征冗余问题,模块引入池化层(Pooling)对特征图进行下采样,有效压缩了输入维度,减少了网络参数量,显著提升计算效率的同时,优化了特征分布的稀疏性。这种稀疏优化通过降低特征自由度(degrees of freedom),减轻了模型的过拟合风险,进一步增强了隐患特征提取的鲁棒性。最终,通过独立特征提取模块的设计与分布式特征融合策略的结合,实现了多源物探数据的高效整合,为隐患识别提供了更强大的特征表达能力和空间分辨率。
式(1)中:Ffused为融合后的特征表示;Fi为第 i 个模态的输入特征;Xi为对应特征的权重系数;Wi为动态融合权重;σ为非线性激活函数;通过引入动态调整机制,网络能够根据不同数据源的重要性自动优化Wi,从而实现特征的高效整合;b为偏置项, 用于调整激活函数的输出,增强模型的表达能力;Concat为拼接操作。
在堤坝隐患识别任务中,MPA衡量模型对不同隐患类别的均衡识别能力,避免因数据不均衡导致的类别偏倚。MPA需80%以上,以确保模型能有效识别各类隐患区域,特别是在稀有隐患(如地下水通道)中,MPA的提升表明模型对小样本类别的敏感性增强。与此同时,交并比(intersection over union, IoU)[29]作为语义分割精度的重要衡量标准,计算预测区域与真实标签区域的交集与并集之比,IoU值越高,说明模型对隐患区域的覆盖准确性越强。其数学表达式为
为提高模型的训练效率和收敛速度,提出一种融合的训练与识别策略,结合自适应矩估计优化器(adaptive moment estimation, Adam)[30]、动态学习率调度和多GPU并行计算等多项技术。这些策略不仅有效提升了模型训练过程中的稳定性,还加速了收敛过程,确保最终模型在堤坝隐患识别中的高效性和准确性,尤其在多类别隐患识别任务中展现出优异的性能。
ZhouJianping, DuXiaohu, ZhouXingbo, et al. Special article: research on world high dams and its future development trends[J]. Journal of Hydroelectric Engineering, 2019, 38(2): 1-14.
[3]
WengerC. Better use and management of levees: reducing flood risk in a changing climate[J]. Environmental Reviews, 2015, 23(2): 240-255.
LiuShanjun, WuLixin, MaoYachun, et al. Intelligent monitoring technology and typical application of open-pit mine slope based on satellite-air-ground collaboration[J]. Journal of China Coal Society, 2020, 45(6): 2265-2276.
[6]
DeringG, MicklethwaiteS, ThieleS T, et al. Insights into dyke emplacement mechanismsusing new high-resolution 3D digital outcrop methods[J]. Journal of the Geological Society, 2019, 176(3): 445-460.
WuAiqing, WuQinghua. Evolution mechanism of embankment hazards and rapid detection of hidden dangers and emergency rescue equipment[J]. Chinese Journal of Geotechnical Engineering, 2022, 44(7): 1310-1328.
[9]
LiS, LiuB, XuX, et al. An overview of ahead geological prospecting in tunneling[J]. Tunnelling and Underground Space Technology, 2017, 63: 69-94.
[10]
JavanF D, SamadzadeganF, ToosiA, et al. Unmanned aerial geophysical remote sensing: a survey[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17: 4992-5015.
XuZhenhao, MaWen, LinPeng, et al. Intelligent lithology recognition based on rock image transfer learning[J]. Journal of Basic Science and Engineering, 2021, 29(5): 1075-1089.
ChenPeishuai, YuanQing, ZhangZiping, et al. Image interpretation method for geological advance prediction in water-rich fracture zone of tunnel based on convolutional neural network[J]. Journal of Basic Science and Engineering, 2022, 30(1): 196-207.
WuHao, ChenYuntao, ZhuZhaohui, et al. Classification prediction of tunnel squeezing deformation based on improved one-dimensional convolutional neural network[J]. Journal of Basic Science and Engineering, 2024, 32(1): 145-159.
LüQingfeng, LiYu, NiuRong, et al. Research on deformation prediction of surrounding rock in special geotechnical tunnels based on deep learning[J]. Journal of Basic Science and Engineering, 2023, 31(6): 1590-1601.
DuJingyi, ChenZhen, ZhangJiawei, et al. Identification and location of leakage hole diameter of underground water supply pipeline in coal mines[J]. Science Technology and Engineering, 2025, 25(8): 3296-3303.
WuZiyan, JiaDawei, WangQi’ang. Building crack identification based on convolutional neural network and regional growth method[J]. Journal of Basic Science and Engineering, 2022, 30(2): 317-327.
ZhangYan, GuoDaojing, ZhangShuguang, et al. Recognition and sensitivity analysis of limestone corrosion degree in karst areas using convolutional neural network[J]. Journal of Basic Science and Engineering, 2023, 31(4): 961-976.
LiuZongjie, HeXilei, ZhangZuhao, et al. Low-order fault identification technology based on 3D U-Net fully convolutional neuralnetwork[J]. Progress in Geophysics, 2021, 36(6): 2519-2530.
HeYilong, WenXiaotao, WangJintao, et al. Fault identification based on 3D U-Net convolutional neural network[J]. Progress in Geophysics, 2022, 37(2): 560-567.
[29]
SiddiqueN, PahedingS, ElkinC P, et al. U-Net and its variants for medical image segmentation: a review of theory and applications[J]. IEEE Access, 2021, 9: 82031-82057.
[30]
CaoH, TianY, LiuY, et al. Water bodyextraction from high spatial resolution remote sensing images based on enhanced U-Net and multi-scale information fusion[J]. Scientific Reports, 2024, 14(1): 16132.
XieZhihui, WangWenshuang, LiuXuefeng. Research on small sample recognition algorithm for fitting images based on improved U-Net[J]. Foreign Electronic Measurement Technology, 2024, 43(2): 51-58.
TangJie, MengTao, HanShengyuan, et al. Fault detection method for seismic data based on multi-resolution U-Net network[J]. Oil Geophysical Prospecting, 2021, 56(3): 436-445.
HuTing, XuBin, WangYongfa, et al. Research on first arrival picking of mine microseismic based on U-Net[J]. Science Technology and Engineering, 2023, 23(16): 6802-6809.
WangNanyang, ShenJianghai. Image segmentation method of core holes based on MSHAM-UNet[J]. Science Technology and Engineering, 2024, 24(24): 10362-10369.
LiuEnliang, YuanJunliang, HuoShoudong, et al. Automatic identification of shale geological outcrop fracture characteristics based on deep learning[J]. Science Technology and Engineering, 2024, 24(30): 12853-12863.
ZhangJun. Research on mine transient electromagnetic synthetic aperture imaging processing system[J]. Progress in Geophysics, 2016, 31(5): 2255-2262.
[47]
王炳和, 李宏昌. 声纳技术的应用及其最新进展[J]. 物理, 2001(8): 491-495.
[48]
WangBinghe, LiHongchang. Application and latest progress of sonar technology[J]. Physics, 2001(8): 491-495.
[49]
MollM A, BairdH S, AnC. Truthing for pixel-accurate segmentation[C]// The 8th IAPR International Workshop on Document Analysis Systems. New York: IEEE, 2008: 379-385.
ZhouWei, LiuYuxiang, LiaoGuangping, et al. Research on siamese network target tracking algorithm combined with intersection over union loss[J]. Journal of System Simulation, 2022, 34(9): 1956-1967.
PanJinye, WangMiaomiao, KanWei, et al. Lithium-ion battery state of charge estimation method based on Adam optimization algorithm and long short-term memory neural networks[J]. Electrical Technology, 2022, 23(4): 25-30.