基于多尺度跃层卷积神经网络的精细车型识别

为解决精细车型识别中特征不具有代表性,且识别准确率低的问题,提出了基于多尺度跃层卷积神经网络(CNN)的车型识别方法。通过多个不同尺度的跃层卷积神经网络,提取适用于精细车型识别的低层局部特征和高层全局特征,并分别训练Softmax分类器。利用自适应方式融合方法,将多个单一尺度跃层卷积神经网络的识别结果进行融合,调整不同网络对识别结果的贡献。实验中车型识别准确率达到97.59%。实验结果表明多尺度跃层卷积神经网络适用于精细的车型识别,并能提高识别的准确率。

Fine Vehicle Recognition Based on the Multiscale Layer-skipping Convolutional Neural Network

In order to solve the problem in vehicle make-and-model recognition that the features are not representative and the recognition accuracy is low, a method based on multiscale layer-skipping convolutional neural network(CNN) is proposed. First, extracting local features and global features by multiscale layer-skipping convolutional kernels, and then train the softmax classifiers. An adaptive fusion method is used to adjust the contribution of different networks. The recognition results of some single scale layer-skipping convolutional neural networks are fused, and then get the final classification models. The recognition accuracy of the model is 97.59%. The experimental results show that the multiscale layer-skipping convolutional neural network is suitable for fine vehicle recognition, and can improve the accuracy of recognition.