改进的YOLOv5图书梯标检测算法  

Improved YOLOv5 for book ladder label detection algorithm

作　　者：杨祥[1] 王华彬 董明刚[2] YANG Xiang;WANG Huabin;DONG Minggang(College of Computer Science and Engineering,Guilin University of Technology,Guilin 541006,China;College of Physics and Electronic Information Engineering,Guilin University of Technology,Guilin 541006,China)

机构地区：[1]桂林理工大学计算机科学与工程学院,广西桂林541006 [2]桂林理工大学物理与电子信息工程学院,广西桂林541006

出　　处：《桂林理工大学学报》2024年第2期330-338,共9页Journal of Guilin University of Technology

基　　金：国家自然科学基金地区项目(61563012)。

摘　　要：对于印刷行业而言,确保每本书的页码、页面正确排序是提高印刷厂生产率的重点。为了解决目前印刷厂对于图书梯标的检测存在漏检、误检的问题,提出了一种改进YOLOv5的梯标检测算法。该算法增加一个检测尺度以解决漏检问题和提高梯标的识别精度,将主干网络中的C3数量减少,使用深度可分离卷积替代颈部网络中的部分卷积,在能够保证精度稳定的同时降低模型的参数量和计算量;使用SPConv替换Resunit结构里的Conv,形成SPConv-C3结构,SPConv-C3结构不仅能够降低模型的参数量,还能提升网络对图书梯标的检测精度;根据深度可分离卷积和SPConv-C3结构设计特征融合网络结构,能够大幅度提升网络模型的特征融合能力。将所改进的检测算法在自建数据集上进行验证,结果表明,改进后的算法Pm,A@0.5∶0.95达到了64.7%,比原YOLOv5s提高7.5百分点,参数量下降了3.16%,能够满足实际需求。For the printing industry,ensuring that the page numbers and pages of each book are correctly sorted is the focus of improving the productivity of printing plants.In order to solve the problem of missed detection and false detection of book ladder label in printing plants,a ladder label detection algorithm to improve YOLOv5 is proposed.The algorithm first adds a detection scale.The purpose is to solve the problem of missed detection and improve the recognition accuracy of the ladder label.Secondly,the number of C3 in the backbone network is reduced,and then the deep separable convolution instead of partial convolution in the neck network,which can ensure stable accuracy and reduce the amount of parameters and calculations of the model.Then SPConv is used to replace the Conv in the Resunit structure to form an SPConv-C3 structure,which can not only reduce the amount of parameters of the model,but also improve the detection accuracy of book ladder label on network.Finally,the feature fusion network structure is designed according to the depth separable convolution and SPConv-C3 structure,which can greatly improve the feature fusion capability of the network model.The improved detection algorithm is experimented on the self-built dataset,and the experimental results show that the improved algorithm Pm,A@0.5∶0.95 reached 64.7%,which is 7.5 percentage point higher than the original YOLOv5s,and the parameter amount is reduced by 3.16%,which can meet the actual needs.

关 键 词：YOLOv5 梯标检测 目标检测 基于拆分的卷积 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]