Neural Network Methodologies and Their Potential Application to Cloud Pattern Recognition.
Abstract
Six artificial neural network models are explored as potential methodologies for the automated interpretation of satellite cloud images. The Multi-layer Perceptron Network is chosen to be the most applicable to the image interpretation problem. A complete, mathematical description of the methodology is presented. The neural network's classification capability is demonstrated using simple geometric patterns and alphabetic characters. A more complex test using GOES infrared imagery shows that the neural network can distinguish 53 of 54 large-scale cloud patterns. An architecture for a complete, automated cloud feature recognition system is proposed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1991
- Accession Number
- ADA239214
Entities
People
- J. E. Peak
Organizations
- Computer Sciences Corporation