首页|隧道掌子面节理卷积神经网络智能识别方法

隧道掌子面节理卷积神经网络智能识别方法

扫码查看
为解决现有隧道掌子面节理裂隙识别方法中存在的识别精度不足、鲁棒性较低以及检测速度较慢等问题,提出了一种名为 Mask Region-convolutional Neural Network-EfficientNet(Mask R-CNN-E)的隧道掌子面节理裂隙识别算法.该算法以Mask R-CNN实例分割算法为基础,采用先进的EfficientNet网络作为主干网络,加强了 Mask R-CNN算法获取特征信息的能力,从而显著提升了识别精度.EfficientNet通过复合缩放方法(Compound Scaling Method)有效地平衡了网络的深度、宽度和分辨率,使得在计算效率和准确率之间达到了最佳平衡.在模型训练过程中,采用了多尺度训练方式和学习率调整策略poly,以增强算法的鲁棒性.为评估算法的性能,以平均精度值Am为测试指标,与传统的Mask R-CNN算法开展了对比试验.此外,采用骨架算法对模型检测输出的节理裂隙掩码进行细化处理,以获取更为精确的节理裂隙量化信息.研究结果表明:改进后的算法在预测框平均精度值(b_Am)和分割平均精度值(s_Am)上分别达到了 0.656和0.436,均显著高于传统方法,表明了其在识别精度上的优势;改进后的Mask R-CNN-E算法在隧道掌子面节理裂隙识别方面有显著提升,尤其在复杂的隧道作业环境中表现出更强的鲁棒性和抗干扰能力;在节理裂隙长度测量方面,算法的误差控制在1.5%~9.8%之间,能够更好地满足工程实际需求.该方法不仅在理论上有较高的精度和鲁棒性,在实际应用中也能提供更为可靠的支持,对于提高隧道工程的安全性和施工效率具有重要意义.
Intelligent Recognition Method of Tunnel Face Joints and Fissures Using Convolutional Neural Network
To address the issues of insufficient recognition accuracy,low robustness,and slow detection speed in existing tunnel face joint and fissure recognition methods,this paper proposes a novel algorithm called mask-region convolutional neural network-EfficientNet(Mask R-CNN-E)based on the Mask R-CNN instance segmentation algorithm for tunnel face joint and fissure recognition.This algorithm incorporates the advanced EfficientNet as the backbone network to enhance the feature extraction capability of Mask R-CNN,thereby significantly improving recognition accuracy.EfficientNet employs a compound scaling method to effectively balance network depth,width,and resolution,achieving an optimal tradeoff between computational efficiency and accuracy.During the model training process,multiscale training and poly-learning rate adjustment strategies were adopted to enhance the robustness of the algorithm.The performance of the algorithm was evaluated using the mean average precision(Am)metric,and comparative experiments were conducted using the traditional Mask R-CNN algorithm.In addition,a skeleton algorithm was employed to refine the joint and fissure mask outputs of the model to obtain more precise quantitative information on joints and fissures.The results show that the improved algorithm achieved a bounding box mean average precision(b_Am)of 0.656 and a segmentation mean average precision(s_Am)of 0.436,with both significantly higher than those of the traditional method,indicating superior recognition accuracy.The improved Mask R-CNN-E algorithm significantly enhances tunnel face joint and fissure recognition,exhibiting stronger robustness and anti-interference capabilities in complex tunnel environments.In terms of joint and fissure length measurements,the algorithmic error was controlled within the range of 1.5%-9.8%,which satisfies engineering requirements.This method not only offers high theoretical accuracy and robustness but also provides more reliable support in practical applications,which is crucial for improving the safety and efficiency of tunnel engineering.

tunnel engineeringtunnel face sketchinstance segmentationjoints and fissuresdeep learning

张运波、雷明锋、肖勇卓、刘光辉、邓兴兴、杨富宇、鲁宝金、李重阳

展开 >

中南大学土木工程学院,湖南长沙 410075

贵州路桥集团有限公司,贵州贵阳 550000

隧道工程 掌子面素描 实例分割 节理裂隙 深度学习

国家重点研发计划项目贵州路桥集团有限公司科技项目贵州省交通运输厅科技项目贵州省交通运输厅科技项目

2023YFB260390GPTJ-18-QJ-012023-122-0082021-122-047

2024

中国公路学报
中国公路学会

中国公路学报

CSTPCD北大核心
影响因子:1.607
ISSN:1001-7372
年,卷(期):2024.37(7)
  • 14