基于YOLOv7-Tiny的车牌及放大号检测研究  

Research on License Plate and Amplification Detection Based on YOLOv7-Tiny

作　　者：陈冠宇 尚雅层[1] CHEN Guan-yu;SHANG Ya-ceng(School of Mechanical and Electrical Engineering,Xi′an University of Technology,Xi′an 710021,China)

机构地区：[1]西安工业大学机电工程学院,西安710021

出　　处：《价值工程》2024年第7期104-106,共3页Value Engineering

基　　金：陕西省重点研发计划项目(2020GY-160)。

摘　　要：由于大型车辆的后车牌容易污损和遮挡,目前车牌号的识别对于这种情况有明显缺陷。提出将车牌放大号和车牌共同检测后再识别,提升车牌识别算法的适用性。本文基于YOLOv7-Tiny检测,算法先后采用更换主干网络和卷积模块实现模型轻量化,通过改进损失函数来提升精度。实验表明,在YOLOv7-Tiny更换Mobilenetv3主干网络、GSConv卷积核和Focal-EIoU后,实现模型体积下降35%,参数量下降37%,运算量下降58%,从而实现一种轻量化的模型。Due to the tendency of large vehicles to have their rear license plates easily stained and obstructed,current license plate recognition based on license plates has obvious shortcomings in this situation.In response to this issue,this article proposes to amplify the license plate detection and license plate detection before recognition,in order to improve the applicability of license plate recognition algorithms.This article adopts YOLOv7-Tiny detection,and successively uses replacing the backbone network and convolutional module to achieve model lightweight,improving the loss function to improve accuracy,and thus achieving a lightweight model.The experiment showed that after replacing the Mobilenetv3 backbone network,GSConv convolutional kernel,and Focal EIoU with YOLOv7 Tiny,the model volume decreased by 35%,parameter quantity decreased by 37%,and computational complexity decreased by 58%with a slight decrease in map value.

关 键 词：放大号 YOLOv7-Tiny Mobilenetv3 GSConv Focal-EioU 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]