Comparing expert systems and neural fuzzy systems for object recognition in map dataset revision
Recognition of objects extracted from remotely sensed imagery requires the matching of object properties with prior stored knowledge. Various properties were used in this study to form the model of a priori knowledge. A GIS (geographical information system) dataset was used to assist the extraction of shape descriptors, reflectance and height above ground characteristics of classes of object including building, road, grassland and tree. Human interpreters are capable of recognizing objects in natural scenes (including aerial photography) that display complex, overlapping composition and representation. Objects extracted from such imagery are inherently fuzzy. In order to perform the recognition task by computer, such uncertainty must be accommodated. Many researchers have used the robustness of neural networks to accomplish such recognition. In this work, we utilized a fuzzy expert system and an adaptive neuro-fuzzy system to train, adapt and recognize objects in three complex aerial scenes.