Machine learning for automatic classification of remotely sensed data

摘要

As more and more remotely sensed data becomes available it is becoming increasinglyharder to analyse it with the more traditional labour intensive, manualmethods. The commonly used techniques, that involve expert evaluation, arewidely acknowledged as providing inconsistent results, at best. We need moregeneral techniques that can adapt to a given situation and that incorporate thestrengths of the traditional methods, human operators and new technologies.The difficulty in interpreting remotely sensed data is that often only a smallamount of data is available for classification. It can be noisy, incomplete or containirrelevant information.Given that the training data may be limited we demonstrate a variety of techniquesfor highlighting information in the available data and how to select the mostrelevant information for a given classification task. We show that more consistentresults between the training data and an entire image can be obtained, and howmisclassification errors can be reduced. Specifically, a new technique for attributeselection in neural networks is demonstrated.Machine learning techniques, in particular, provide us with a means of automatingclassification using training data from a variety of data sources, including remotelysensed data and expert knowledge.A classification framework is presented in this thesis that can be used with anyclassifier and any available data. While this was developed in the context ofvegetation mapping from remotely sensed data using machine learning classifiers,it is a general technique that can be applied to any domain. The emphasis ofthe applicability for this framework being domains that have inadequate trainingdata available.