Evaluation of Object-Oriented Rule-Based and Example-Based Classification Methods to Extract the Mountain Roads Using Satellite Images

With the advent of the high spatial resolution satellites in recent years, feature extraction has attracted the attention of many researchers. Road has a great importance in infrastructure, transportation and management. The extraction of urban road network has been the subject of numerous studies. Because of its low contrast, complex geometric structure and far away from crowded city centers, mountain road extraction from satellite images has been less studied. In this study to extract mountain roads QuickBird satellite imagery was used. In order to increase the spatial resolution of multispectral images, panchromatic and multispectral bands were fused using novel Esri method. Rod extraction was performed using advanced object-oriented classification methods and the results were evaluated. The overall accuracy for Object-oriented rule-based and Example-based classification methods was 93% and 92% respectively. Considering the results, road's central axis was extracted from the region generated by the Object-oriented rule-based method. Final accuracy of central axis was 87% and RMSE 1.1 pixel comparing to the reference data. Due to the Topographic complexity of the area and spatial resolution of the images, the results showed that advanced Object-oriented classification method has a good capability to extract homogeneous terrestrial objects especially the roads.