An intercomparison of artificial intelligence approaches for polar scene identification

• Main Authors: Tovinkere, V. R., Penaloza, M., Logar, A., Lee, J., Weger, R. C., Berendes, T. A., Welch, R. M.
• Format: Other/Unknown Material
• Language: unknown
• Published: 1993
• Subjects: 47; Arctic; Sea ice
• Online Access: http://ntrs.nasa.gov/search.jsp?R=19930048364

The following six different artificial-intelligence (AI) approaches to polar scene identification are examined: (1) a feed forward back propagation neural network, (2) a probabilistic neural network, (3) a hybrid neural network, (4) a 'don't care' feed forward perception model, (5) a 'don't care' feed forward back propagation neural network, and (6) a fuzzy logic based expert system. The ten classes into which six AVHRR local-coverage arctic scenes were classified were: water, solid sea ice, broken sea ice, snow-covered mountains, land, stratus over ice, stratus over water, cirrus over water, cumulus over water, and multilayer cloudiness. It was found that 'don't care' back propagation neural network produced the highest accuracies. This approach has also low CPU requirement.