基于二级改进LeNet-5的交通标志识别算法

发布时间：2018-03-19 17:19

本文选题：卷积神经网络　切入点：交通标志　出处：《陕西师范大学学报(自然科学版)》2017年02期 　论文类型：期刊论文

【摘要】：以真实场景中拍摄的交通标志图像数据集GTSRB为研究对象,将卷积神经网络与支持向量机相结合,提出一种基于二级改进LeNet-5的交通标志识别算法。该算法首先根据识别系统的实时性要求,对原始LeNet-5结构进行改进;然后用裁剪、灰度化、图像增强和尺寸归一化等操作对原始图像进行预处理,得到32×32的感兴趣区域;接下来,利用数据集GTSRB训练出一个二级改进LeNet-5,其中第一级改进LeNet-5将感兴趣区域中包含的交通标志粗分为6类,第二级改进LeNet-5对粗分类结果进行细分类,识别出交通标志所属的最终类别。实验结果表明,基于二级改进LeNet-5交通标志识别算法因网络模型能够提取交通标志的多尺度特征,识别正确率可达91.76%。
[Abstract]:Taking the real scene traffic sign image data set GTSRB as the research object, the convolution neural network and the support vector machine are combined. A traffic sign recognition algorithm based on two-level improved LeNet-5 is proposed. Firstly, the original LeNet-5 structure is improved according to the real-time requirements of the recognition system, and then the original LeNet-5 structure is clipped and grayscale. Image enhancement and size normalization operations preprocess the original image to obtain 32 脳 32 regions of interest. A second-level improved LeNet-5 is trained by using the data set GTSRB, in which the first improved LeNet-5 divides the traffic signs contained in the region of interest into six categories, and the second improved LeNet-5 subdivides the results of the rough classification. The experimental results show that the traffic sign recognition algorithm based on two-level improved LeNet-5 can extract the multi-scale features of traffic signs, and the recognition accuracy can reach 91.76.
【作者单位】：陕西师范大学计算机科学学院;现代教学技术教育部重点实验室;陕西省语音与图像信息处理重点实验室;
【基金】：国家自然科学基金(61501287,61501286) 陕西省重点实验室开放共享项目(SAIIP201202) 陕西省自然科学基础研究计划(2015JQ6208)
【分类号】：U463.6;U495;TP391.41

【参考文献】