基于多尺度卷积神经网络的交通标志识别方法

为了提升交通标志自动识别的精度,提出一种基于多尺度CNN的交通标志识别方法(TSR -MSCNN算法).该方法采用三阶段卷积神经网络,融合了低阶、中阶和高阶3种不同尺度的特征,并串联了多个小卷积层用以代替单个较大卷积层.通过对全连接层的神经元个数、Dropout参数、卷积核尺寸等网络超参数进行选比实验,获得了最佳的网络超参数.利用德国交通标志基准数据库(GTSRB)对不同算法进行测试表明,本文提出的算法在较小的网络参数量下能够有效提取交通标志特征,获取的识别准确率达到99.76%,且显著优于传统卷积神经网络方法和多尺度特征方法的识别准确率,因此本文算法在图像识别领域有良好的应用价值.

Traffic sign recognition method based on multi - scale convolutional neural network

In order to improve the accuracy of automatic traffic sign recognition, we propose a traffic sign recognition algorithm based on multi -scale CNN. This method uses a three -stage convolutional neural network to fuse features of three different scales: low -order, medium -order, and high -order; and concatenates multiple small convolutional layers to replace a single larger convolution layer. In addition, the network hyperparameters such as the number of neurons in the fully connected layer, the dropout parameters, and the sizes of the convolution kernels are investigated and compared to obtain the best hyperparameter set. Different algorithms are tested on the German traffic sign recogninion benchmark(GTSRB). Experimental results show that the proposed algorithm in this paper can effectively extract traffic sign features and obtain recognition accuracy of 99.76% under a small amount of network parameters, which is obviously superior to the traditional convolutional neural network method and multi -scale feature recognition method. Therefore, the algorithm proposed in this paper has good usability in the field of image recognition.