Analyzing Fuzzy Logic, Logistic-Decision Tree, and Neural Network Classification for Extracting Subzonal Land Uses from Remote Sensing Imagery

摘要

Lack of elaborate land use (LU) information has forced city planners and modelers touse large aggregated zones in their models, and consequently accept undesirable approximationsand errors in their analyses and planning workflow. This paper presents and compares theperformances of three classification techniques developed and designed for LU extractionthrough a hybrid geographic information (GI)/remote sensing (RS) expert system. The hybridsystem is designed to classify urban subzones, in this case, dissemination blocks (DBs), into pureLUs using very high resolution (VHR) aerial imagery and several GIS data. The LUclassification techniques developed in this study, which are the core of the proposed expertsystem, includes Fuzzy-Decision Tree, Logistic-Decision Tree, and Artificial Neural Network(ANN). Several types of GIS data, including building footprint, street network, digital propertymap (DPM), and dissemination block (DB) zones from the study area, City of Fredericton,Canada, are fused into the LU classification expert systems to discover correlation/associationrules related to urban LU classes. Morphological properties of a number of selected DBs, whichcontain different types of pure LUs, are derived from GIS/RS data. Zonal properties, includingvegetation ratio, soil ratio, building ratio, street ratio, parking ratio, mean building area, meanbuilding perimeter, mean building compactness, average height of buildings, and sum of soil andparking ratio, are used as potential indicators in the classification process. Based on the accuracyassessment, the testing results of the three classification techniques indicate that Logistic-Decision Tree has the best performance with an overall accuracy of 95.2%. Continuingdevelopment of the proposed GI/RS expert system will have an important implication to currentmodeling process by providing up-to-date and much more detailed land use information.