摘要
在图像分类识别中,为了获得更高的分类精确度,需要对图片提取更精确和更能表现图片语义信息的特征,深度学习已成为特征提取最常用的方法。提出一种改进的深度卷积神经网络的图片分类模型。通过从网络架构和内部结构两方面对经典的深度神经网络AlexNet的改进和优化,进一步提升特征的表达能力。通过在全连接层引入极限学习机,不仅提高了网络的分类能力和分类时间,而且使得该结构具有更优的数据处理能力。通过在两个标准数据集上的一系列对比实验,分析了不同的优化方法在不同情况下的作用,并证明了该网络结构的有效性。
In image classification and recognition, more accurate and better image semantic information features need to be extracted to obtain higher classification accuracy.Deep learning becomes the most common method for feature extraction.We proposed a picture classification model based on improved deep convolutional neural network in this paper.The classic deep neural network AlexNet was improved and optimized from two aspects： network architecture and internal structure, to further enhance the expressive ability of features.The extreme learning machine was introduced at the full connection layer, which gave the network a better classification capability and time, and made this structure more capable of data processing.Through comparison experiments on two standard data sets, it analyzes the effect of different optimization methods under different conditions and proves the effectiveness of the network structure.
作者
闫河
王鹏
董莺艳
罗成
李焕
Yan He;Wang Peng;Dong Yingyan;Luo Cheng;Li Huan(College of Computer Science,Chongqing University of Technology,Chongqing 401320,China;College of Liangjiang Artifical Intelligence,Chongqing University of Technology,Chongqing 401147,China)
出处
《计算机应用与软件》
北大核心
2018年第12期193-198,共6页
Computer Applications and Software
基金
国家自然科学基金面上项目(61173184)
重庆市自然科学基金项目(cstc2018jcyjAX0694)
关键词
卷积神经网络
极限学习机
图片识别
池化
特征提取
Convolutional neural network
Extreme learning machine
Image recognition
Pooling
Feature extraction
