Large Area Road Extraction Using High Resolution SAR Images

Synthetic aperture radar is an active microwave imaging radar; it can work in any weather during the whole day. It has the working characteristics of not affected by light and climate conditions during all-weather and whole day. It plays a cue role in many fields, such as environmental protection, disaster monitoring, ocean observations, resources exploration, precision agriculture, geological mapping and the government public. With the improvement of SAR image resolution, the amount of image data in SAR imaging system is increase, and the size of a single image is increasing, however, the technology of data processing is difficult to meet the requirements of real-time and effective treatment. As a typical artificial way, the rode is the main part of the modern transportation system. It becomes more and more important in the extraction of SAR images. However, with the increasing of resolution, all other kinds of interference will be enlarged. The road extraction task becomes complex and difficult, it have many factors, such as shadow of the both sides, the presence of vehicles on the road, the complexity and diversity of the environmental background.In order to use the SAR image data more efficiently, on the basic of keeping the efficiency and reliability of SAR image target automatic interpretation the reinforced earth observation to strengthen the ability of earth observation and useful information gathering. This paper focuses on the problem of road network recognition from high resolution remote sensing images. The study include some key technology such as image filtering, edge detection and regions of interest detection, multi-resolution feature extraction. And on this basis build a road targets detection and identification method. This method meet the requirement of target extraction is simple, fast and strong adaptability. To overcome the existing road detection algorithm are pointed and the poor robustness.For the purpose of improving the utilization rate of SAR image, increasing the problem of interference information, huge scene of the image and timeconsuming targets searching, we propose a visual saliency detection based ROI extraction. Different scales' potential targets take different approaches. Long straight structure targets in low resolution images using a line segment length density figure significantly down from the visual model significantly. The areas of more than threshold or a percentage of the long straight line segments or parallel lines are contrast as ROI, and it can find the candidate large linear targets in original image quickly. Finally, using the human visual significant mechanism to determine the ROI to reduce the dealing data further. This method can quickly determine the potential location of large-format image target and improve the efficiency of extraction.According to need to extract the road network, focusing on the parallel line features of road in high-resolution SAR images, and then establishes a new model of parallel lines and uses this model in SAR image. Finally, we can get the connection criteria based on heuristic line primitive. Eventually, we can get the overall program of road extraction in the high-resolution SAR images. Experimental results show that the model can extract features from the parallel line for the high-resolution SAR images quickly.At the end of this paper, we give out the experimental tests of the given out algorithm. And analysis of the results qualitative and quantitative. Experimental results show that the algorithm which proposed in this paper can detect the highresolution SAR images road network better, it have some applicability.