Accession Number:

ADA238986

Title:

Application of Neural Networks to Large-Scale Cloud Pattern Recognition

Descriptive Note:

Final rept.,

Corporate Author:

COMPUTER SCIENCES CORP MONTEREY CA

Personal Author(s):

• Peak, J. E.

Report Date:

1991-03-01

Pagination or Media Count:

45.0

Abstract:

A Multi-layer Perceptron Neural Network methodology is used to classify eight types of large-scale cloud patterns. The data are taken from GOES-W visible images from Oct. 1-Dec. 31, 1983. Large-scale features are previously identified by a human expert to provide a data set for supervised learning. Discriminant Analysis is used to reduce the set of network inputs and as a comparison classification methodology. In three different tests, the neural network technique classifies the cases with consistently higher accuracy than Discriminant Analysis. The problem of image segmentation is addressed in a preliminary test of the Hierarchical Stepwise Optimization algorithm.

Descriptors:

• *CLASSIFICATION
• *ARTIFICIAL INTELLIGENCE
• *PATTERN RECOGNITION
• *CLOUDS
• OPTIMIZATION
• DISCRIMINATE ANALYSIS
• ACCURACY
• PATTERNS
• SEGMENTED
• METHODOLOGY
• IMAGE PROCESSING
• ALGORITHMS
• DATA BASES
• COMPARISON

Subject Categories:

• Meteorology
• Cybernetics

Distribution Statement:

APPROVED FOR PUBLIC RELEASE