ANN BASED CLASSIFIER FOR SATELLITE IMAGES

Abstract

In the past frw decades remote sensing imagery utility has proved a powetf iii technology for monitoring earth's swface at global regional and local scales. As the sensor technolog)' to capture the images advances, the volume of re/note sensing images also continues to grow. Geospatial data is widely used now in the fields such as government, ,nilitary, industry, research etc. The use is in these fields require accurate classification of terrain propertie.s in areas where there are ongoing operations or locations of interest. There are many applications in analysis and classification qf satellite images such as in geology, soil science, detection o/ water areas, vegetation etc.Accuraie land cover classification from satellite imagery are important/or all the abm'e stated applications. Traditional classifiers have been used to extract LUC information s but their application is limited to onh' a few datasets. On the other hand ANN and SVM provide a nonlinear and accurate ways of' classifying satellite images without relying on their statistical in/orination. The ANN as trained hvfeatures extracted from different segments of the image.The work done in this thesis investigates the possibility of comnbining object based and pixel based class (fier or increasing the accuracy of classification.