摘要
为解决传统Vgg-16网络模型较大,模型损失率较高和准确率较低收敛速度较慢的问题,在传统Vgg-16网络基础上结合YOLO网络模型对传统VggNet进行改进,在VggNet的基础上利用全局平均池化层代替全连接层进行输出,减少了网络模型参数,并在轻量型YOLO网络的基础上增强了对小目标的识别率。首先通过调节图片光照、图像增强以及归一化处理等操作对收集到的交通标志图片进行预处理,建立完整的国内交通标志数据集后,将定位与识别分成两部分进行。利用改进的轻量型YOLO网络对图像中的目标进行定位,并利用对小目标识别准确率较高的改进后的VggNet对定位后的目标进行分类识别,并将改进过后的训练模型应用到道路交通标志的识别上。从实验结果中得出,文中所提出的网络模型对于小目标的检测比YOLO网络增加2.66%,且网络参数较传统VggNet显著降低。
In order to solve the problem that the traditional Vgg-16 network needs to manually select parameters such as learning rate,and the network model is large,and the convergence speed of model loss rate and accuracy is slow,the traditional VggNet is improved based on the traditional Vgg-16 network combined with YOLO network model.Based on VggNet,the global average pooling layer is used to replace the full connection layer for output,to reduce the network model parameters.Based on the lightweight YOLO network,the recognition efficiency of small targets is enhanced.Firstly,the collected traffic sign images are preprocessed by adjusting the image illumination,image enhancement and normalization processing.After establishing a complete domestic traffic sign data set,the positioning and recognition are divided into two parts.The improved lightweight YOLO network is used to locate the targets in the image,and the improved VggNet with high accuracy of small target recognition is used to classify and recognize the located targets,and the improved training model is applied to the recognition of road traffic signs.The experimental results show that the network model proposed in this paper increases the detection of small targets by 2.66%compared with YOLO network,and the network parameters are significantly lower than traditional VggNet.
作者
李飞
李昊
喻自烘
LI Fei;LI Hao;YU Zihong(School of Information and Engineering,Shenyang University of Technology,Liaoning Shenyang 110870,China)
出处
《机械设计与制造》
北大核心
2025年第12期86-89,共4页
Machinery Design & Manufacture
基金
面向膝上截肢者融合智能下肢假肢的新型外骨骼机器人关键技术研究(61803272)。
