基于全局-局部卷积神经网络的阻生牙分类  

Global and local based convolutional neural networks for impacted tooth classification

作　　者：杨鎏 舒祥波 YANG Liu;SHU Xiangbo(School of Information and Software Engineering, University of Electronic Science and Technology of China, Chengdu Sichuan 610054, China;West China Hospital of Stomatology, Sichuan University, Chengdu Sichuan 610041, China;School of Computer Science and Engineering, Nanjing University Science and Technology, Nanjing Jiangsu 210094, China)

机构地区：[1]电子科技大学信息与软件工程学院,成都610054 [2]四川大学华西口腔医院,成都610041 [3]南京理工大学计算机科学与工程学院,南京210094

出　　处：《计算机应用》2019年第A01期250-253,共4页journal of Computer Applications

摘　　要：为了解决因牙齿位于口腔的局部区域,且阻生牙与正常牙之间的视觉差异性非常微小导致的观察口腔CT图片可能出现的误判的问题,提出了一种新的全局局部卷积神经网络模型(GL-CNN)。首先从全局图像与局部区域分别学习口腔CT图像中能够带有阻生牙判别信息的特征描述,然后将这两种学习到的特征进行整合来训练一个支持向量机(SVM)分类器。为了验证GL-CNN的性能,在收集的口腔CT图像数据集上用提出的GL-CNN方法与基准方法进行阻生牙分类对比实验,实验结果表明了GL-CNN方法具有更高的分类精度。A new Global and Local based Convolutional Neural Network (GL-CNN) was proposed to solve the problem of misjudgment in the observation of oral Computed Tomography (CT) images due to the fact that the teeth are located in the local area of the oral cavity and the visual difference between impacted teeth and normal teeth is very small.Firstly, feature descriptions of impacted teeth in oral CT images were learned from global images and local regions respectively. Then the two learned features were integrated to train a Support Vector Machine (SVM) classifier. The proposed GL-CNN method was compared with the benchmark method in the collected oral CT image data set. The experimental results show that the proposed GL-CNN method has higher classification accuracy.

关 键 词：阻生牙分类 卷积神经网络 图像分类 深度学习 支持向量机分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]