||
本文为布拉格捷克理工大学(作者:Nikita Tishin)的学士论文,共63页。
近年来,卷积神经网络(CNNs)已成为图像识别领域的一个热门选择。它们为表征学习提供了一个框架,允许在没有任何手动特征提取的情况下进行原始输入处理。本文的第一个目的是对CNNs进行理论研究和理解。第二个目标是评价和比较几种CNN在图像分类中的应用。实验部分包括两个基准问题,即MNIST数据库和CIFAR-10,以及由脑电图(EEG)信号生成的递归图组成的医学数据集。特别地,CNNs被用于对从精神健康者和患有精神疾病的人接收到的EEG数据进行分类。
In recent years Convolutional NeuralNetworks (CNNs) have became a most popular choice for image recognitionproblems. They provide a framework for representation learning, which allowsfor raw input processing without any manual feature engineering. The first goalof this thesis is a theoretical study and comprehension of CNNs. The secondgoal is evaluation and comparison of several CNN applied to the imageclassification. Experimental part includes two benchmark problems, i.e. MNISTdatabase and CIFAR-10, as well as medical dataset consisting of recurrenceplots created from the Electroencephalography (EEG) signals. In particular,CNNs are used to classify EEG data received from mentally healthy persons andpersons having a mental illness.
更多精彩文章请关注公众号:
Archiver|手机版|科学网 ( 京ICP备07017567号-12 )
GMT+8, 2024-9-27 10:07
Powered by ScienceNet.cn
Copyright © 2007- 中国科学报社