摘要
为进一步提高混凝土桥梁裂缝识别的准确率,并提高识别效率,基于一阶段目标检测算法中的YOLOv5算法和注意力机制模块,提出了YOLOv5_CBCA算法。在CBS(Convolution,Batch Normalization,SiLU)模块中融入注意力机制CBAM(convolutional block attention module)模块以减少降采样对特征提取的影响,骨干网络尾部添加CA(coordinate attention)模块降低图像背景的影响,从而提高目标定位的准确性。通过消融实验、对比实验,验证了YOLOv5_CBCA算法中改进模块的有效性。通过对双龙堡大桥、钟家大桥等混凝土桥梁裂缝图片进行检测,证明了YOLOv5_CBCA算法在提高准确率的同时具备更好的抗干扰能力,体现了在混凝土桥梁裂缝检测方面的优越性,为一阶段目标检测算法在混凝土桥梁裂缝识别工作中的应用提供了参考。
To further enhance the accuracy of crack detection in concrete bridges,as well as to improve detection efficiency,this study proposes the YOLOv5_CBCA algorithm,based on the YOLOv5 algorithm within the one-stage object detection framework and an attention mechanism module.By integrating the convolutional block attention module(CBAM)into the CBS(Convolution,Batch Normalization,SiLU)module,the impact of downsampling on feature extraction is reduced.Additionally,the inclusion of the CA(coordinate attention)module at the tail end of the backbone network diminishes the effect of the image background,thereby increasing the precision of target localization.The effectiveness of the improved modules within the YOLOv5_CBCA algorithm is validated through ablation studies and comparative experiments.The application of this algorithm to crack images from concrete bridges,such as the Shuanglongbao and Zhongjia bridges,demonstrates its higher accuracy and better anti-interference capability,showcasing its superiority in concrete bridge crack detection.This provides a reference for the application of one-stage object detection algorithms in the identification of concrete bridge cracks.
作者
黄可原
赵毅
胡楠
曹建秋
向阳开
Huang Keyuan;Zhao Yi;Hu Nan;Cao Jianqiu;Xiang Yangkai(School of Civil Engineering,Chongqing Jiaotong University,Chongqi 400074,China;School of Materials Science and Engineering,Chongqing Jiaotong University,Chongqi 400074,China;School of Information Science and Engineering,Chongqing Jiaotong University,Chongqi 400074,China)
出处
《科技通报》
2024年第9期71-76,共6页
Bulletin of Science and Technology
基金
重庆市研究生联合培养基地建设项目(JDLHPYJD2021011)。
