基于卷积神经网络的车型颜色综合识别

Synthetic Recognition of Vehicle Color Based on Convolution Neural Network

在高速公路复杂因素干扰的情况下,存在车型误判和车辆颜色变化较大不易于识别的问题。文中利用深度学习Caffe框架中的LeNet、AlexNet、GoogLeNet3种网络模型对车型和颜色进行综合识别,得出3种网络模型下车型颜色的综合识别率,再与支持向量机(SVM)进行比较。实验结果表明,深度学习Caffe框架下3种模型的识别率相较支持向量机(SVM)的方法得到了大幅提高,且超过90%。其中Caffe框架下的GoogLeNet网络模型准确率可达95%以上,效果明显。

In the case of complex factors of highway interference,there is a problem that the vehicle misjudgment and the vehicle color change are not easy to be identified. This article uses the three network models of LeNet,AlexNet and GoogLeNet in the framework of deep learning to comprehensively identify the models and colors,and obtains the comprehensive recognition rate of models and colors under three kinds of network modelss,and then carries on the support vector machine（ SVM） Comparison. The experimental results show that the three model recognition rates under the deep learning Caffe framework are significantly higher than those of the support vector machine（ SVM）,and more than 90%. Which Caffe framework under the GoogLeNet network model accuracy rate of up to95% or more,the effect is obvious.