| 基于卷积神经网络的图像分类算法综述 |
| |
| 引用本文： | 杨真真,匡楠,范露,康彬.基于卷积神经网络的图像分类算法综述[J].信号处理,2018,34(12):1474-1489. |
| |
| 作者姓名： | 杨真真 匡楠 范露 康彬 |
| |
| 作者单位： | 南京邮电大学通信与网络技术国家工程研究中心 |
| |
| 基金项目： | 国家自然科学基金(61501251,11671004,61271335,61271240);中国博士后科学基金(2018M632326);通信与网络技术国家工程研究中心开放课题(TXKY17010);江苏省自然科学基金青年基金(BK20170915);江苏省高校自然科学面上项目(17KJD510005);南京邮电大学引进人才科研启动基金(NY214191, NY216023)
|
| |
| 摘 要： | 随着大数据的到来以及计算能力的提高，深度学习(Deep Learning, DL)席卷全球。传统的图像分类方法难以处理庞大的图像数据以及无法满足人们对图像分类精度和速度上的要求，基于卷积神经网络(Convolutional Neural Network, CNN)的图像分类方法冲破了传统图像分类方法的瓶颈，成为目前图像分类的主流算法，如何有效利用卷积神经网络来进行图像分类成为国内外计算机视觉领域研究的热点。本文在对卷积神经网络进行系统的研究并且深入研究卷积神经网络在图像处理中的应用后，给出了基于卷积神经网络的图像分类所采用的主流结构模型、优缺点、时间/空间复杂度、模型训练过程中可能遇到的问题和相应的解决方案，与此同时也对基于深度学习的图像分类拓展模型的生成式对抗网络和胶囊网络进行介绍；然后通过仿真实验验证了在图像分类精度上，基于卷积神经网络的图像分类方法优于传统图像分类方法，同时综合比较了目前较为流行的卷积神经网络模型之间的性能差异并进一步验证了各种模型的优缺点；最后对于过拟合问题、数据集构建方法、生成式对抗网络及胶囊网络性能进行相关实验及分析。
|
| 关 键 词： | 卷积神经网络 图像分类 深度学习 生成式对抗网络 胶囊网络 |
| 收稿时间： | 2018-07-27 |
Review of Image Classification Algorithms Based on Convolutional Neural Networks |
| |
| Affiliation: | National Engineering Research Center of Communications and Networking, Nanjing University of Posts and? TelecommunicationsSchool of Science, Nanjing University of Posts and Telecommunications |
| |
| Abstract: | With the arrival of big data and the improvement of computing power, deep learning (DL) has swept the world. The traditional image classification method is difficult to process huge image data and cannot meet the requirements of image classification accuracy and speed. The image classification method based on convolutional neural network (CNN) breaks through the bottleneck of traditional image classification methods. It has become the mainstream image classification algorithm. Therefore, how to effectively use the convolutional neural network to classify images has become a hot topic in the field of computer vision at home and abroad. In this paper, after systematic research on convolutional neural networks and indepth study of the application of convolutional neural networks in image processing, the mainstream structural models and their advantages and disadvantages, time/space complexity are presented. Problems that may be encountered in the process of model training used in image classification based on convolutional neural networks and corresponding solutions are also given. At the same time, it also introduces the generative adversarial network and capsule network of image classification extension model based on deep learning. Then, the simulation results show that the image classification method based on convolutional neural network is superior to the traditional image classification method in image classification accuracy. At the same time, the performance differences between the current popular convolutional neural network models are compared and the advantages and disadvantages are further verified. Finally, Experiments and explanations on over-fitting problems, dataset construction methods, generative adversarial network and capsule network performance are given. |
| |
| Keywords: | |
|
| 点击此处可从《信号处理》浏览原始摘要信息 |
|
点击此处可从《信号处理》下载全文 |
|
