一种基于残差网络改进的牙齿颜色分类模型

A Tooth Color Classification Model Improved by Residual Network

针对传统牙齿比色方法准确率低和效率低等问题,提出一种基于残差网络改进的牙齿颜色分类模型。该模型通过融合多层卷积结果以及引入压缩与激励注意力机制模块的方式,使网络能学习到更多的图像颜色特征。基于典型牙齿所建数据集进行颜色分类实验,在该数据集上对文中模型与GoogleNet、MobileNet-V1、ResNet-34和ResNet-50等模型进行颜色分类预测结果比较。实验结果表明,文中模型优于传统模型,预测分类准确度达到91.16%,有效提高了牙齿颜色分类准确率和效率。

This paper proposes an improved tooth color classification model based on residual network to increase the accuracy and efficiency of traditional tooth colorimetric methods.This model enables the network to learn more image color features by fusing multilayer convolutional results and by introducing Squeeze-and-Excitation(SE)attention mechanism module.Color classification experiments are conducted on a typical teeth dataset,on which the color classification prediction results of the proposed model are compared with those of GoogleNet,MobileNet-V1,ResNet-34 and ResNet-50.The experimental results show that the proposed model is better than the traditional models,and the prediction classification accuracy reaches 91.16%,which effectively improves the accuracy and efficiency of tooth color classification.