基于双通道卷积神经网络的多标签图像标注  

Multi-label image annotation based on double channel convolution neural network

作　　者：陈立潮[1] 武晨燕 曹建芳 潘理虎[1] 张英俊[1] CHEN Li-chao;WU Chen-yan;CAO Jian-fang;PAN Li-hu;ZHANG Ying-jun(School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China;Department of Computer Science and Technology,Xinzhou Teachers University,Xinzhou 034000,China)

机构地区：[1]太原科技大学计算机科学与技术学院,山西太原030024 [2]忻州师范学院计算机科学与技术系,山西忻州034000

出　　处：《计算机工程与设计》2019年第12期3601-3607,共7页Computer Engineering and Design

基　　金：山西省中科院科技合作基金项目(20141101001);“十二五”山西省科技重大专项基金项目(20121101001);山西省科技攻关基金项目(20141039);山西省重点研发计划基金项目(201603D121031)

摘　　要：针对图像语义标注中存在的训练样本不均衡导致低频标注词标注准确率低的问题,提出一种双通道卷积神经网络模型(double channel convolution neural network,DCCNN)。其中一个通道是为训练低频样本设立的,以此提高低频样本在整个模型中所占比重,另一个通道用于训练全部的训练集。在标注过程中把两个通道的输出进行融合,对所需标注的标注词共同做出决策。在Pascal VOC2012标准数据集上对模型进行验证,实验结果表明,DCCNN模型相对于卷积神经网络(convolution neural network,CNN)无论是对低频标注词的标注准确率还是效率都有很大的提升,验证了该模型的有效性。Aiming at the problem that the training samples in the image semantic annotation are not balanced,resulting in low accuracy of low-frequency labeling,a dual channel convolution neural network was proposed.One of the channels was set to train low-frequency samples to increase the proportion of low-frequency samples in the entire model,and the other was used to train all training sets.In the labeling process,the outputs of the two channels were fused to make a decision on the labeling words of the required labels.By verifying the model on the Pascal VOC2012 standard dataset,the experimental results show that the DCCNN model has significant improvement in the accuracy and efficiency of the low-frequency annotation words compared to the convolutional neural network,thus verifying the validity of the model.

关 键 词：图像标注 卷积神经网络 样本不均衡 多标签 双通道卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]