Newland, F.T., Tatnall, A.R.L. and Brown, M.
Neurofuzzy Extraction of Wind Data from Remotely Sensed Images.
Proc. 3rd Int. Wind Workshop
Full text not available from this repository.
Object- and feature-based neurofuzzy techniques for processing METEOSAT images have been developed for generation of height assigned wind data. Three basic types of cloud object motion have been identified, and a fuzzy system has been used to determine the degree of each motion type within image regions. Preliminary results are given for a series of METEOSAT infrared images. Parameters suitable for quantifying each motion type have been implemented. Previous work has successfully applied a neural network approach to matching one of the parameter sets over time, for generating cloud motion wind vectors. A neurofuzzy approach to generation and combination of wind vectors from all the motion parameter sets defined has been considered. In addition to the increase in generated wind vector data, the proposed system has the advantage of a core fuzzy rule base which allows for easier wind vector quality control, as the network's decision processes can be more easily visualised and interpreted within the context of cloud motion analysis. It also offers a more flexible interface with which to adapt the object matching criteria for tracking features, for example to incorporate new object motion types or motion analysis parameters, and could be used to incorporate expert knowledge of specific meteorological phenomena.
Actions (login required)