Urban Land-cover Mapping and Change Detection with Radarsat Sar Data Using Neural Network and Rule-based Classifiers

This paper presents a new approach to extract urban landuse/land-cover information from high-resolution radar satellite data. Fivedate RADARSAT fine-beam SAR images over the rural-urban fringe of the Greater Toronto Area were acquired during May to August in 2002. One scene of Landsat TM imagery was acquired in 1988 for change detection. The major landuse/land-cover classes were high-density built-up areas, low-density built-up areas, roads, forests, parks, golf courses, water and four types of agricultural crops (soybeans, corn, winter wheat/rye and pasture). The proposed approach to classify SAR images consisted of three steps: 1) image segmentation, 2) feature selection and object-based neural network classification, 3) rule set development to improve classification accuracy. Post-classification change detections were then performed using the final classification result of RADARSAT SAR images and the classification result of Landsat TM imagery. The results showed that the proposed approach achieved very good classification accuracy (overall: 87.9%; kappa: 0.867). The change detection procedure was able to identify the areas of significant changes, for example, new built-up areas, even though the overall accuracy of the change detection was not high.