文章基本信息

标题：Study of hierarchical learning and properties of convolution layer using sign language recognition model
本地全文：下载
作者：Sagaya Mary J ; Nachamai M ; M.Vijayakumar 等
期刊名称：Ilköğretim Online/Elementary Education Online
印刷版ISSN：1305-3515
电子版ISSN：1305-3515
出版年度：2021
卷号：20
期号：5
页码：1118-1127
DOI：10.17051/ilkonline.2021.05.121
语种：English
出版社：Öğretmen Eğitimi Akademisi
摘要：Convolution Neural Network (CNN) as a technique improves research minds to overcome the challenges of handcrafted feature extraction and classification. CNN be a part of the representation learning methods in deep learning architecture to discover the representation needed for detection and classification automatically. So far this technique has been thought as “black boxes”, meaning that their inner working principles are mysterious and inscrutable. In order to understand the internal behavior of CNN, a model is developed on sign language recognition with 99.81%, 94.69%, 92.60% accuracy in train, test, and validation. While developing a model the inner principles of automatic feature extraction and the unique properties of convolution operations available in hierarchical CNN architecture are also learned. CNN is a multilayered network leading to feature learning and classification, it is necessary to understand how the features are learned from each layer and how it is transformed and fed into the next higher level layers without any human interventions..
关键词：deep learning;convolution neural network;kernel;sparse connection;parameter sharing