From: Hugh Lewis Date: Thu, 11 Aug 94 12:57:23 BST Subject: Cloud Classification Application of Neural Networks To Satellite Imagery--Cloud Classification and Quantification of Cloud Radiative Forcing. Hugh Lewis*, Dr Adrian Tatnall* and Dr Martin Brown** ---------- Department of Aeronautics and Astronautics, The University of Southampton, Highfield, Southampton SO17 1BJ, UK. *Astronautics Research Group. **Advanced Systems Research Group. We are exploring the use of Neural Networks and other intelligent control systems for classifying clouds and quantifying their radiative forcing effects. To date we have used a Multi-Layer Perceptron, trained using back propagation, to classify a limited amount of surfaces: High Cloud, Low Cloud, Land and Sea. The network was trained on raw-data samples from a two-channel (visible and infrared) image, then tested on the entire image. We have had promising results from this simple system, and hope to extend the number of classification categories and explore the use of statistical features, self-organising networks and unsupervised learning, and the possible introduction of a neuro-fuzzy system. Images are being kindly provided by the UK Meteorological Office. Please contact me if you would like to know more about this research, or if you would like a draft- copy of our paper (Application of Neural Networks To Satellite Imagery) which describes the work in a little more detail. Hugh Lewis ------------------------ hglewis@uk.ac.soton.aero Tel: +44 703 59 4896 Fax: +44 703 59 3058