基于注意力机制和迁移学习的轮胎花纹分类

Tire Classification Based on Attention Mechanism and Transfer Learning

随着国家建设“平安城市”规划的提出,交通安全和城市治安管理变得尤为重要。而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。本文提出一种基于注意力机制和迁移学习的轮胎花纹分类方法。通过自适应直方图均衡化对于输入的轮胎图片进行数据增强,使轮胎花纹纹理与背景部分对比更加显著。通过迁移学习,减少训练时间,通过改进ResNet50模型,融合通道空间注意力机制提高了模型对于轮胎花纹纹理的特征提取能力。通过实验证明,该模型对于轮胎花纹分类,其最高精度可以达到96.41%,表明有较强的泛化能力。

With the national construction of"Safe City"plan put forward,traffic safety and urban public security management has become particularly important.Tire pattern image classification plays an important role in traffic accidents and criminal investigation and evidence collection.This paper proposes a tire pattern classification method based on convolutional neural network and transfer learning.Data enhancement is performed on the input tire image through adaptive histogram equalization,so that the contrast between the tire pattern texture and the normal part is more significant.Through transfer learning,the training time is reduced.By improving the ResNet50 model and integrating the channel spatial attention mechanism,the feature extraction ability of the model for tire tread texture is improved.Experiments show that the model can achieve a maximum accuracy of 96.41%for tire pattern classification,indicating that it has a strong generalization ability.