A hybrid histogram/neural network classifier for creating global cloud masks
Abstract. A histogram method is used to identify the features which are used to classify the scene elements, and a hierarchical neural network with 'don`t care` nodes is used to perform the actual classification. Although the ultimate goal is to produce a classifier capable of detecting clouds anywhere in the world, this preliminary system is limited to classifications in three distinct ecosystems: the polar region, desert terrain, and scenes of biomass burning in South America. Classification accuracies of 97.5 per cent for desert data, 98 per cent for polar data, and 99.2 per cent for biomass burning data were achieved. More importantly, manual analysis of classified scenes which had not been previously seen by the classifier show 'very good` to 'excellent` correspondence between the computer generated classifications and the human expert classifications.
No Reference information available - sign in for access.
No Citation information available - sign in for access.
No Supplementary Data.