Visidyne proposes to develop an expert, neural net based, system for classifying cloud types based on texture analysis in satellite imagery using geometric measures as network inputs. In Phase I Visidyne proposes to investigate the utility of a variety of network inputs based on the geometry and characteristic spatial scales of cloud texture in satellite imagery. These measures will be compared with the more traditional statistical ones for a sampling of different cloud types in NOAA satellite images. If a set of geometric measures with good correlation to visually typed cloud images is found, then development of a neural net based expert system for classifying cloud types will be proposed for Phase II. The primary benefit of Phase I will be the evaluation of the utility of geometric as opposed to statistical measures for use in cloud typing based on satellite image texture analysis. If Phase I is successful then candidate sets of geometric measures will have been identified for evaluation as inputs for a neural network based expert system for cloud typing in satellite imagery.
Keywords: Cloud Type Nowcasting Satellite Imagery Geometric Measures Statistical Measures Neural Networks
