|
| 基于改进YOLOv7算法的隧道衬砌裂缝智能识别研究 |
| Intelligent Identification of Tunnel Lining Cracks Based on Improved YOLOv7 |
| 投稿时间：2024-02-06 修订日期：2024-03-19 |
| DOI： |
| 中文关键词: 隧道衬砌裂缝 深度学习 目标检测 GSConv BiFormer Wise-IoU v3损失 |
| 英文关键词: Tunnel lining cracks Deep learning Target detection GSConv BiFormer Wise-IoU v3 Loss |
| 基金项目:国家自然科学基金资助项目(52078211)湖南省交通运输厅科技进步与创新项目(202308) |
|
| 摘要点击次数: 80 |
| 全文下载次数: 0 |
| 中文摘要: |
| 隧道衬砌裂缝的快速识别和准确分类,为专业人员高效及时的修复隧道衬砌的破损提供便利。为了克服传统人工目测法的不便,本文提出针对隧道衬砌裂缝特征的改进YOLOv7算法模型Tunnel Lining Crack-YOLO(TLC-YOLO)。本文对比了四类骨干网络对隧道衬砌裂缝的检测效果,认为复杂环境下裂缝检测存在强背景干扰、训练样本质量不平衡等问题,通过TLC-YOLO模型使用轻量级卷积GSConv和slim-neck架构,嵌入动态稀疏注意力模块BiFormer,加强通道信息传输来提高模型的实时反应速度和检测精度,可实现更灵活的计算分配和内容感知。使用Wise-IoU v3作为坐标回归损失函数,为较好的训练样本分配梯度,抑制较差的训练实例,以此来提高模型的泛化能力。研究结果表明,通过隧道裂缝数据集训练之后,与YOLOv7相比,在多组实验中TLC-YOLO 模型对隧道裂缝同时提高了检测结果的准确率、召回率、F1 值以及mAP@0.5 值,证明了TLC-YOLO对隧道衬砌裂缝有更好的检测和分类能力。 |
| 英文摘要: |
| The rapid identification and accurate classification of tunnel lining cracks facilitates professionals to repair tunnel lining damage in an efficient and timely manner. In order to overcome the inconvenience of the traditional manual visual inspection method, this paper proposes Tunnel Lining Crack-YOLO (TLC-YOLO), an improved YOLOv7 algorithm model for tunnel lining crack characteristics. This paper compares the detection effect of four types of backbone networks on tunnel lining cracks, and concludes that crack detection in complex environments has problems such as strong background interference and imbalance in the quality of training samples, etc. By using lightweight convolutional GSConv and slim-neck architectures in the TLC-YOLO model, and by embedding a dynamic sparse attention module, BiFormer, and by enhancing the transmission of channel information, we can improve the real-time response speed and detection accuracy of the model, which can be realized. reaction speed and detection accuracy, enabling more flexible computation allocation and content awareness. Wise-IoU v3 is used as a coordinate regression loss function to improve the generalization ability of the model by assigning gradients to better training samples and suppressing poorer training instances. The results show that after training through the tunnel crack dataset, the TLC-YOLO model simultaneously improves the accuracy, recall, and F1 values, and mAP@0.5 values of the detection results for tunnel crack lesions in multiple sets of experiments compared to YOLOv7, proving that TLC-YOLO has a better ability to detect and classify tunnel lining cracks. |
|
View Fulltext
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
