中文摘要：

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%。

英文摘要：

Focusing on GTSRB dataset acquired in real world, a traffic sign recognition algorithm based on the 2-level improved LeNet-5 is proposed,which combines convolutional neural networks with support vector machines. With the consideration of the requirement of real-time recognition, the traditional network structure of LeNet-5 is improved first. After GTSRB dataset images were cropped and converted to grayscale images, their brightness and size are normalized to 32 X 32 im- ages. Next, a 2-level improved LeNet-5 is trained with GTSRB dataset, where the first level cat- egorized traffic signs to 6 categories with the improved LeNet-5, and the second level improved LeNet-5 provide with the final category. Experimental results show that the proposed algorithm could provide with a correct recognition ratio 91. 76%, since the multi-scale features could be fully analyzed with 2-level improved LeNet-5.