Neural network system for cloud classification from satellite images

Highly accurate, automated cloud detection and classification methods are essential for processing multispectral meteorological satellite in an operational environment and providing data for meteorological and climatological studies. They help to discover hazardous meteorological phenomena such as hail storms developing on tops of clouds, hurricanes and cyclones. Weather prediction and rainfall estimation systems are enhanced substantially by having access to information about cloud cover distribution and being able to trace changes in meteorological conditions. This paper reports on a research project, sponsored by the British Natural Resources Institute, to investigate the problem of cloud feature extraction, detection and classification. The paper presents a novel architecture and implementation of a hybrid two-stage neural network system for cloud detection and classification from satellite imagery