基于CNN、SVM和迁移学习的轮胎花纹分类

Tire tread pattern image classification based on SVM,CNN and transfer learning

轮胎花纹图像分类在交通事故及刑侦破案取证中具有重要的作用。为了准确地分类轮胎花纹图像,提出了一种基于卷积神经网络(convelutional neural network,CNN)、支持向量机(support vector machine,SVM)和迁移学习的轮胎花纹分类算法。首先对辅助数据库ImageNet进行CNN训练得到初始CNN模型;其次,基于迁移学习思想,利用轮胎花纹图像数据库对初始CNN模型的分类层进行微调训练,得到用于轮胎花纹图像分类的CNN模型;最后,从所得CNN模型的第二个全连接层提取输出的4 096维特征,用该特征对轮胎花纹图像进行基于SVM的图像分类。使用轮胎花纹图像数据库进行分类实验,结果表明,提出算法的分类精度达到93.1%。说明提出算法能够提高轮胎花纹图像的分类准确率。

Tire tread pattern image classification plays an important role in traffic accidents and criminal investigation.In order to accurately classify the tire tread image,an effective tire tread pattern classification algorithm is proposed based on convelutional neural network（CNN）,support vector machine（SVM）and transfer learning.Firstly,CNN is trained on ImageNet,to get the initial CNN model.Secondly,based on the idea of migration learning,the tire pattern image database is used to finetune the classification layer of the initial model,and the model is then used for tire pattern image classification.Finally,the output 4096 dimension features are extracted from the second full connection layer of the tire pattern model,and this feature is used to classify tire pattern images based on SVM.The classification experiment is carried out on the database of tire pattern images,and the classification accuracy of the proposed algorithm can reach 93.1 percent,indicating that the proposed algorithm can classify tire pattern images more accurately.