Study Of The Road Extraction From GF-1 Remote Sensing Images

Using the domestic high resolution earth observation data for independent analysis is an important goal of China for building the spatial information application system. Extracting the road information from remote sensing images is an economic and feasible way and it is a research hotspot in remote sensing application. Limited to the technical level, road extraction from remote sensing images still need to much manual intervention.This paper uses GF-1 satellite images in Loess Plateau Region as experimental data to research on road extraction including surfaced roads and dirt roads. The experimental images are not of high resolution and contented of complex topography, thus the roads have low width and can be easily affected by surrounding terrain, leading to broken roads. This paper uses two remote sensing indexs on multi spectral image to obtain low vegetation coverage region and non-water area as the road potential region. Based on object oriented thought and panchromatic image in the road potential area, we use multi-scale segmentation method and edge detection result as assistance to segment the panchromatic image into objects.Then using the road objects features, fuzzy classification to extract road, then using mathematical morphology to optimize road’s morphology. Based on the fast marching method, in this paper, we develop an interactive tool to extract the missing road parts in the above classification result. This method can be extended to the Loess Plateau Area from GF-1 images road extraction.