基于轻量型卷积神经网络的交通标志识别
投稿时间:2020-08-06  修订日期:2020-09-16  点此下载全文
引用本文:
摘要点击次数: 13
全文下载次数: 0
作者单位E-mail
龙曼仪 长沙理工大学 2679553242@qq.com 
李茂军 长沙理工大学 2679553242@qq.com 
基金项目:国家自然科学基金
中文摘要:针对卷积神经网络在交通标志识别实时性不好,对设备硬件要求过高的缺点,提出了一种具有实时性,高精度的基于轻量型卷积神经网络的改进网络。一方面引入深度可分离卷积和激活函数Mish,加快网络的训练和识别速度,降低对硬件设备的要求;另一方面通过对网络架构及层次的改进,同时合理的改变卷积核的大小和数目,加强图片特征的表达与传递。在BelgiumTSC交通标志数据集上的实验结果表明,改进的网络明显提高了网络训练速度,同时识别精度也略高于原网络,验证了本文方法的有效性。通过与其他模型相比,该模型能够更快速准确完成交通标志识别任务,验证了该方法的可行性。
中文关键词:卷积神经网络  交通标识  图像增强  深度可分离卷积  激活函数
 
Traffic sign recognition based on lightweight neural network
Abstract:Aiming at the shortcomings of convolutional neural network in the recognition of traffic signs that the real-time performance is not good, and the equipment hardware requirements are too high, a real-time and high-precision improved network based on lightweight convolutional neural network is proposed. Separate convolution and activation function Mish, speed up the network training and recognition speed, reduce the requirements for hardware equipment; on the other hand, through the improvement of the network architecture and level, while reasonably changing the size and number of convolution kernels, the expression of image features and transfer. The experimental results on the BelgiumTSC traffic sign dataset show that the improved network significantly increases the network training speed, and the recognition accuracy is slightly higher than the original network, which verifies the effectiveness of the method in this paper. Compared with other models, this model can complete the task of traffic sign recognition more quickly and accurately, which verifies the feasibility of this method.
keywords:Convolutional neural network  Traffic sign  Image processing  Depth separable convolution  Activation function
查看全文   查看/发表评论   下载pdf阅读器