Neocognitron of a Neural Network for Seismic Pattern Recognition

Kou-Yuan Huang*, J. Y. Liaw

*Corresponding author for this work

Research output: Contribution to conferencePaperpeer-review

Abstract

This paper presents the design of an artificial neural network, called a "neocognitron", for seismic pattern recognition. The neocognitron was proposed by Fukushima and is quite important in handwritten numerals. A seismic pattern recognition system which works with the mechanism of the neocognitron is discussed in this paper to demonstrate the ability of the neocognitron which can overcome the difficulty with the seismic patterns being distorted in shape, changed in size, and shifted in position. The system has been trained using a supervised classification. The neocognitron is a hierarchical multilayered network consisting of a cascade of many layers of neuron-like cells of the analog type, and has variable connections between the cells in adjoining layers. It can acquire the ability to recognize patterns by training. During the training process, "teacher" presents a set of training patterns and points out which cells should be the seed cells for each training patterns. Training of each subsequent layer does not begin until the training of the previous layer has finished. After finishing the process of learning, pattern recognition is performed on the basis of similarity in shape between patterns, and not affected by the deformation. Shift invariance, scale invariance and noise tolerance can be achieved through a proper choice of design parameter. In the neocognitron, local features of the input pattern are extracted by the cells of lower stage, and gradually integrated into more global feature. Finally, each cell of the highest stage integrates all the information of the input pattern, and responds only to one pattern. In our experiment, the neocognitron has successfully recognized the distorted input patterns, and correct classification could be maintained with up to 10% of the input pixels corrupted.

Original languageEnglish
Pages26-29
Number of pages4
StatePublished - 1 Jan 1992
Event1992 Society of Exploration Geophysicists Annual Meeting - New Orleans, United States
Duration: 25 Oct 199229 Oct 1992

Conference

Conference1992 Society of Exploration Geophysicists Annual Meeting
CountryUnited States
CityNew Orleans
Period25/10/9229/10/92

Fingerprint Dive into the research topics of 'Neocognitron of a Neural Network for Seismic Pattern Recognition'. Together they form a unique fingerprint.

Cite this