Improved YOLOv5 for book ladder label detection algorithm
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 SP-Conv-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 im-proved 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.
国家科技期刊平台
NSTL
万方数据
