摘要
提出一种基于深度特征融合的图像分类方法。通过不同的深度学习预训练网络来获取图像的高层语义特征;采用Weighted Discriminant Correlation Analysis(WDCA)方法提取其转换矩阵及其融合矩阵;通过支持向量机分类器进行分类识别。在Caltech 256标准数据库上的实验表明,该方法不但能够有效地优化整合不同的深度特征,而且能够有效地降低特征的冗余信息,从而使融合后的特征具有很强的鉴别能力和低维特点。
This paper presents an image classification method based on deep feature fusion.The high-level semantic features of images were obtained through different deep learning pre-training networks;the weighted discriminant correlation analysis(WDCA)method was used to extract the transformation matrix and its fusion matrix;the support vector machine classifier was used for classification and recognition.Experiments on the Caltech 256 standard database show that our method not only can effectively optimize the integration of different deep features,but also effectively reduce the redundant information of features,so that the fused features have strong discriminating ability and low dimensional characteristics.
作者
蔡志锋
袁宝华
刘广海
Cai Zhifeng;Yuan Baohua;Liu Guanghai(College of Computer Science and Engineering,Sanjiang University,Nanjing 210000,Jiangsu,China;Department of Computer Science and Technology,Taizhou Institute of Sci.&Tech.,Nanjing University of Science and Technology,Taizhou 225300,Jiangsu,China;College of Computer Science and Information Engineering,Guangxi Normal University,Guilin 541004,Guangxi,China)
出处
《计算机应用与软件》
北大核心
2020年第10期175-179,243,共6页
Computer Applications and Software
基金
国家自然科学基金项目(61866005)。
关键词
深度特征
图像分类
特征融合
Deep feature
Image classification
Feature fusion
