AUTOMATIC ROAD FEATURE EXTRACTION FROM HIGH RESOLUTION SATELLITE IMAGES USING LVQ NEURAL NETWORKS
Jayan Wijesingha · Asian Journal of Geoinformatics · 2013
Accurate and up to date road data is very crucial for effective urban planning, infrastructure development, navigation applications, military purposes and for updating topographic GIS databases. Spatial resolutions of remote sensing images are rapidly increasing. It enables a good source for extracting road information. This research has developed a novel approach for sub-urban and rural road feature extraction from high resolution images. An Artificial Neural Network (ANN) based method has introduced by the study with self-organizing supervised learning neural network and compare its performance with typical pattern recognition neural network. The developed ANN models were trained and applied to a World View – II satellite’s panchromatic image. The resulted raster binary images were evaluated using a completeness assessment compared with a digitized reference road layer. Final extracted road layer could be used to update a GIS database. Furthermore, the study has utilized the best learning parameters in developing LVQ and Patternnet NN to extract road from high resolution images. The successful results show the possibility of using the approach for automatic road feature extraction from high resolution images for updating GIS databases.