Ning Yu

Document Type

Thesis - University Access Only

Award Date


Degree Name

Master of Science (MS)

Department / School

Computer Science

First Advisor

Bin Cong


One of the major areas where neural networks are often applied is imaging classification. In this application, a small region from an image is extracted as a pattern that consists of integer numbers. The neural network can either be trained to remember set patterns, or it can be operated in reverse by associating each image area with a known pattern. This paper proposes the approach of using neural networks to classify multispectral remote-sensing images which are the satellite imagery collected at the EROS Data Center in Sioux Falls. Several methodologies are developed for representing data and training a network. Two typical neural network models, the backpropagation and Monte Carlo neural networks, are explored to solve this problem. A hybrid training method is also developed. Criteria for alternating the two algorithms are studied. In addition, a partially connected network is proposed to investigate the relation of the connectivity in a neural network and the accuracy of the classification.

Library of Congress Subject Headings

Remote-sensing images -- Classification
Neural networks (Computer science)




South Dakota State University



Rights Statement

In Copyright