基于改进YOLOv7算法的智慧城市复杂场景车牌识别方法研究  

Research on License Plate Recognition Method for Complex Scenes in Smart Cities Based on Improved YOLOv7 Algorithm

作　　者：鲁娟[1] 卢英杰 李明海[2] Lu Juan;Lu Yingjie;Li Minghai(Design and Research Institute Co.,Ltd.of Xi'an University of Architecture and Technology,Xi'an 710055,China;School of Information and Control Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China)

机构地区：[1]西安建筑科技大学设计研究总院有限公司,西安710055 [2]西安建筑科技大学信息与控制工程学院,西安710055

出　　处：《绿色建造与智能建筑》2024年第5期128-134,共7页GREEN CONSTRUCTION AND INTELLIGENT BUILDING

摘　　要：针对复杂环境下,现有的车牌检测算法存在漏检及模型参数量过大等问题,提出一种改进YOLOv7的中文车牌检测方法。首先,YOLOv7的骨干网络Backbone使用轻量级卷积GhostConv减少模型训练参数。其次,引入CBAM注意力机制,提升小目标车牌的特征提取能力。然后,Head部分使用可感知特征图的空间信息的CoordConv替换部分卷积,提升模型在背景干扰较大场景下的检测性能。最后,将损失函数CIoU更换为NWD,提高算法在训练过程中的收敛速度。实验结果表明,改进算法提高了模型检测精度的同时可大幅减少模型参数和计算量,实现了复杂场景下中文车牌检测的高精度检测性能。Aiming at the problems of leakage detection and excessive amount of model parameters of existing license plate detection algorithms in complex environments,a Chinese license plate detection method with improved YOLOv7 is proposed.Firstly,the backbone network of YOLOv7 uses lightweight conv olutional GhostConv to reduce the model training parameters.Secondly,CBAM attention mechanism is introduced to improve the feature extraction ability of small target license plate.Then,Head partly replaces part of the convolution using CoordConv,which can perceive the spatial information of the feature map.to improve the detection performance of the model in the scene with large background interference.Finally,the loss function CIoU is replaced with NWD to improve the convergence speed of the algorithm during training.The experimental results show that the improved algorithm improves the detection accuracy of the model while significantly reducing the model parameters and computational volume.and realizes the high-precision detection performance of Chinese license plate detection in complex scenes.

关 键 词：复杂场景 车牌检测 YOLOv7 注意力机制 损失函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]