遥感与地理信息系统学报

抽象的

Evaluation of the Full Lambda and SVM Methods Capability to ExtractRoads from Digital Images

Abdollahi A, Bakhtiari HRR and Nejad PM

Automatic extraction of information on ground using photogrammetry and remote sensing requires the formulation of human data and image data, so that, it must include all the content of the image. Complex structure of the various objects in the image leads to the challenges for doing this. So, choose the type of digital data and a good way to extract the desired effect is important in mapping accuracy. This study has investigated the semi-automated method of extraction of various types, including straight, spiral, intersection, urban and non-urban roads from satellite and aerial images. Data used included UltraCam aerial image, Worldview satellite image of non-urban area with a resolution of 0.5 m, and Quick-Bird images of Tehran province with a resolution of 0.61 m. In the proposed method, upon performing image segmentation by using Full lambda method, image classification has been done using SVM algorithm, and morphological operations are used to improve the quality of discover ways and remove noise and cover gaps. For images in which Full lambda method has high accuracy in image segmentation, therefore, the accuracy of the image classification is increased and extraction of the road from it has been done better. The average overall accuracy of over 81 percent and the average accuracy Kappa coefficient more than 78 percent in the image classification into two classes of road and non-road indicates very good capability of the system introduced for semi-automatic extraction of different roads.