Research On Methods Of Road Extraction In SAR Image

The Synthetic Aperture Radar (SAR) has been widely used in military andcivilian areas, because of its weather-independent, all time and strong penetratingcapability. Many countries have devoted a great deal of attention and capital into SAR,so that this technology has made enormous progress in previous decades.Comparatively speaking, SAR image understanding and interpretation has made littleprogress. As an important target in SAR image, the road recognition and extractionhas great theoretical and practical significance, which attract more and more people toresearch. Under the support of“Eleventh Five-Year”National Defense pre-researchproject, this paper focuses on how to, following the order of road elements detection,road segments extraction and road segments connection, extract road targets inhigh-resolution SAR images effectually and accurately, as well as its performanceevaluation method. The main work is as follows:(1) Road elements detection in high-resolution SAR images. As the basis of roadextraction process, the road elements detection immediately determines accuracy andintegrity of final extraction result. Considering that road targets behave like lineobjects, the detection of road elements is essentially the specific application of edgedetection methods on road extraction problem. Traditional SAR image edge detectionalgorithms have some disadvantages: ROA’s edge location is not accurate enough,and low sensitivity of thin roads, while Cross-Correlation algorithm has pooranti-noise capability and higher false alarm rate. On the base of previous research, afusion algorithm is proposed in this chapter, which combines those two algorithms.Compare of experimental results and analysis of performance indexes show that, thisnew algorithm is able to make expected progress on detection rate and false alarm rate.At the same time, the fusion algorithm is compared with Duda algorithm and theresult turns out its powerful ability at eliminating false information. So, thisalgorithm will be used in post-treatment.(2) Road segments extraction. Because of using edge detection at the fist part, thestructural description of road elements set is necessary，which can be devided into twoparts: road segments extraction and feature description. This paper introduces twomethods of road segments extraction: Phase Grouping and Hough Transform, and analyze performance of both through experiments. Aiming at appearance of fracturein road segments caused by direction quantization error and disadvantages of thosetwo methods, a new method is proposed here. This new method combines bothPhase Grouping and Local Hough Transform, so that it can extract road segmentsmore effectively. At last, response average vale ave，direction and normalizedlengthlengthnormare used to describe road segments’feature.(3) Road segments connection. After road elements detection and segmentsextraction, we need to connect road segments globally. This paper uses GA to connectroad segments because of its global optimizing ability. Based on those work above, anew method of automatic road extraction is proposed, as well as two performanceindexes: detection rate and false detection rate. Experiment results prove that the newmethod is able to extract road targets in high-resolution SAR images efficiently andaccuately, and achieves fully automatic road extracting in high-resolution SARimages.(4) Considering the requirements of practical project, this paper implements anotherroad extraction algorithm that based on Hough Transform. This method uses Dudaalgorithm to detect road elements, global Hough Transform to extract and connectroad segments together, and finally modifies the result with Human-ComputerInteraction. But it fails at completeness and semi-automatic processing. Otherwise,after study of VC/Matlab mixed programming, road extraction algorism are packagedas COM components and transplanted into SAR image processing platform. Mainfunctional instances was shown at last part.