摘要
桥梁裂缝具有连续性和不具备个体完整性的特点,将会导致目标检测过程出现大量检测框重叠,造成过检。该算法基于YOLOv4目标检测算法框架,对YOLOv4网络的检测头进行分支改进,输出特征预测热力图,使检测结果具有更好的回归和判断能力;使用热力图辅助网络进行训练,引入Dice系数(Dice coefficient)损失对YOLOv4损失函数进行重新定义,有效的改善了过检问题。研究还发现,在数据集训练过程中使用随机裁剪联合马赛克数据增强策略,提高了网络模型在实际检测场景中的泛化能力。实验结果表明,在相同的迭代次数和数据集下,改进的YOLOv4网络相比于未改进的YOLOv4网络检测精度和速度均有了明显的提升。
Bridge cracks have the characteristics of continuity and incompleteness,which will lead to a large number of overlapping detection frames in the target detection process,resulting in over-detection.The algorithm is based on the YOLOv4 target detection algorithm framework,improves the detection head of the YOLOv4 network,and outputs the feature prediction heat map,so that the detection results have better regression and judgment capabilities;the heat map is used to assist the network for training,and the Dice coefficient loss redefines the YOLOv4 loss function,which effectively improves the over-checking problem.The study also found that using random cropping combined with mosaic data augmentation strategy during data set training improves the generalization ability of the network model in actual detection scenarios.The experimental results show that under the same number of iterations and data sets,the improved YOLOv4 network has significantly improved detection accuracy and speed compared to the unimproved YOLOv4 network.
作者
廖延娜
宋超
Liao Yanna;Song Chao(College of Electronic Engineering,Xi'an University of Posts and Telecommunications,Xi'an 710121,China)
出处
《国外电子测量技术》
北大核心
2022年第4期112-118,共7页
Foreign Electronic Measurement Technology
基金
陕西省重点研发计划—国际科技合作计划项目(2020KW—001)
西安邮电大学研究生创新基金重大项目(CXJJLD202005)资助
作者简介
廖延娜,硕士,副教授,主要研究方向为电路与系统,信号与信息处理,图像处理;通信作者:宋超,硕士研究生,主要研究方向为机器学习、深度学习与图像处理。E-mail:1205914498@qq.com
