Accession Number:

ADA247916

Title:

A Probabilistic Neural Network Approach to Cloud Classification

Descriptive Note:

Final rept.,

Corporate Author:

NAVAL OCEANOGRAPHIC AND ATMOSPHERIC RESEARCH LAB MONTEREY CA

Personal Author(s):

• Bankert, R. L.
• Rabindra, P.
• Sengupta, S. K.

Report Date:

1991-10-01

Pagination or Media Count:

35.0

Abstract:

Automated satellite image interpretation would be useful in many forecasting operations. One aspects of that interpretation, cloud classification, is examined. Ten classes, composed of low, middle, high, and precipitation cloud types plus clear, are used as output nodes in a Probabilistic Neural Network PNN approach to classification of data using four Advanced Very High Resolution Radiometer AVHRR subscenes. Input to the neural network consists of 12 features that include a mixture of spectral, textural, and physical measures. There measures are selected, using a feature selection routine, from a collection of over 200 features. An overall accuracy of 85.25 is the result. Four classes have agreement of 90 or better. The two classes with the poorest accuracies were presented to the classifier with the smallest sample sizes. An increase in the number of samples should increase the accuracy of the classifier.

Descriptors:

• *IMAGE PROCESSING
• CLOUDS
• NEURAL NETS
• AUTOMATION
• PHYSICAL PROPERTIES
• PROBABILITY
• FORECASTING
• ACCURACY
• NODES
• PRECIPITATION
• ARTIFICIAL SATELLITES
• SAMPLING
• CLASSIFICATION
• SELECTION
• OUTPUT

Subject Categories:

• Meteorology

Distribution Statement:

APPROVED FOR PUBLIC RELEASE