基于迁移学习及特征融合的轮胎花纹图像分类

Tread pattern image classification with feature fusion based on transfer learning

为提高图像分类性能,解决因训练数据不足导致卷积神经网络模型过拟合的问题,提出一种基于迁移学习和特征融合的轮胎花纹图像分类算法。将HSV颜色直方图、GIST描述子与方向梯度直方图结合作为轮胎图像低层特征;将迁移学习引入卷积神经网络模型训练中,通过轮胎图像数据集对预训练模型参数微调,获得适用于轮胎花纹图像的新模型,提取全连接层特征作为图像高层特征;将低层和高层特征融合作为轮胎图像最终特征用于训练SVM分类器,实现高效分类。实验结果表明了所提算法的有效性。

To improve image classification performance and relieve the over-fitting problem in convolutional neural network (CNN) model training due to the lack of large scale training data,an effective tread pattern classification algorithm with feature fusion based on transfer learning was proposed.A multi-view low-level feature was designed which was the combination of HSV color histogram,GIST descriptor and histogram of oriented gradient (HOG).Transfer learning was introduced into model trai- ning.The parameters of a pre-trained model were fine-tuned using tread pattern image data,and a new model for the task of tread pattern classification was produced.Features extracted from the fully-connected layer of the new model were used as high-level features of the tread pattern images.Fusion of the low-level feature and high-level feature formed the final feature of the tread pattern image,to train SVM classifier for image classification.The outstanding performance of the proposed algorithm is demonstrated by experimental results.