Images received from satellites have became a great source of information about our environment. This is raw information that needs experts to make the most of it, but there are not many experts and the work is too much. The solution to this problem is the compilation of human experience into automatic systems that could do the same work. We depict here the structure for a knowledge based system capable of taking the place of human experts when it is properly trained. This structure has been used to build an automatic recognition system that process AVHRR images from NOAA satellites to detect and locate ocean phenomena of interest like upwellings, eddies and island wakes. The model covers every phase of the process from the source image, once it is corrected and geocoded, to the final features map. In the most delicate phase of the process pipeline, artificial neural nets and rule-based expert systems are used in a parallel redundant way so results can be validated by comparing the outcome of both subsystems. The automatic knowledge driven image processing system has been trained with ubiquitous and localized information and has proved his qualities with images of Canary Island, Mediterranean Sea and Cantabric and Portuguese coasts.
展开▼
