AUTOMATIC RECOGNITION OF OCEAN STRUCTURES FROM SATELLITE IMAGES BY MEANS OFNEURAL NETS AND EXPERT SYSTEMS

摘要

Images received from satellites have became a greatrnsource of information about our environment. This isrnraw information that needs experts to make the most ofrnit, but there are not many experts and the work is toornmuch. The solution to this problem is the compilation ofrnhuman experience into automatic systems that could dornthe same work.rnWe depict here the structure for a knowledge basedrnsystem capable of taking the place of human expertsrnwhen it is properly trained. This structure has been usedrnto build an automatic recognition system that processrnAVHRR images from NOAA satellites to detect andrnlocate ocean phenomena of interest like upwellings,rneddies and island wakes. The model covers every phasernof the process from the source image, once it isrncorrected and geocoded, to the final features map. In thernmost delicate phase of the process pipeline, artificialrnneural nets and rule-based expert systems are used in arnparallel redundant way so results can be validated byrncomparing the outcome of both subsystems.rnThe automatic knowledge driven image processingrnsystem has been trained with ubiquitous and localizedrninformation and has proved his qualities with images ofrnCanary Island, Mediterranean Sea and Cantabric andrnPortuguese coasts.