Urban landuse/land-cover mapping with high-resolution SAR imagery by integrating support vector machines into object-based analysis

摘要

This paper investigates the capability of high-resolution SAR data for urban landuse/land-cover mapping by integrating support vector machines (SVMs) into object-based analysis. Five-date RADARSAT fine-beam C-HH SAR images with a pixel spacing of 6.25 meter were acquired over the rural-urban fringe of the Great Toronto Area (GTA) during May to August in 2002. First, the SAR images were segmented using multi-resolution segmentation algorithm and two segmentation levels were created. Next, a range of spectral, shape and texture features were selected and calculated for all image objects on both levels. The objects on the lower level then inherited features of their super objects. In this way, the objects on the lower level received detailed descriptions about their neighbours and contexts. Finally, SVM classifiers were used to classify the image objects on the lower level based on the selected features. For training the SVM, sample image objects on the lower level were used. One-against-one approach was chosen to apply SVM to multi-class classification of SAR images in this research. The results show that the proposed method can achieve a high accuracy for the classification of high-resolution SAR images over urban areas.