http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Performance Evaluation on 3-D object recognition using a restricted neural network
Miyanaga, Yoshikazu,Motoyoshi, Katsuhito,Tochinai, Koji 대한전자공학회 1996 APCCAS:Asia Pacific Conference on Circuits And Sys Vol.1 No.1
This report introduces a new approach on a recognition system of three dimensional, i.e., 3D, objects. The proposed system is based on a restricted multi-layered neural network. For the performance evaluation of this network, the experiment in which some similar objects are used for recognition is demonstrated in this report.
ADAPTIVE RECOGNITION OF HAND-WRITTEN KANJI CHARACTERS USING SELF-ORGANIZED NEURAL NETWORK
Miyanaga, Yoshikazu,Tochinai, Koji,Kondo, Masanori,Hayashi, Masato 대한전자공학회 1994 ISPACS:Intelligent Signal Processing and Communica Vol.1 No.1
This port introduces an image recognition system for hand-written Kanji characters. The method is based on a self-organized neural network and a single layer perception network. The self-organized neural network is used for adaptive clustering. The single perception network is used for recognition. It is well known that a large amount of time is required in the training by a mullti-layered perception when some cluster distributions have complicated structures. However, since only the simplest perception is applied in this proposed system, a quite short time is enough to learn training data. The reason why multi-layered perception is not required to recognize data in this system is based on the use of a self-organized network. The self-organized network can change a complicated structure of cluster distribution to a simple structure without the loss of information. Thus, it can be shown that the simple perception is enough to recognize even nonlinear characteristic distribution.
Dynamics Learning Network with Structured Recurrent Modules
Miyanaga, Yoshikazu,Tochinai, Koji,Li, Yisheng 대한전자공학회 1994 ISPACS:Intelligent Signal Processing and Communica Vol.1 No.1
In this report, a local connected recurrent neural network, and a new learning algorithm are proposed. The network which has the ability to memorize and regenerate complex dynamics is constructed by adaptive oscillating modules. This module consists of two simply neuron nodes with recurrent connections. In the new learning algorithm, each module can be trained in independently with suitable sped for given input data. The network's size is also adaptively determined in learning process. Finally, some simulation results are demonstrated to verify the effectiveness of the proposed network structure and the learning algorithm.