The Study, Based On Remote Sensing Images Of The Road Extraction Algorithm

With the rapid development of remote sensing technology, currently, high-resolution remote sensing images are widely used to update based geographic data, survey land resources and detect the changes of using land and so on。In remote sensing images, road information is not only a basic geographic information data, but also it is a target reference and clues to extract other surface features, it is of great significance to correctly extract roads for in-depth application of high resolution remote sensing images。In high resolution remote sensing images, one of the methods of road extraction is firstly using a variety of remote sensing image segmentation methods to segment road target information, and then further extract road objects。Image segmentation is a technology and process that the image is divided into characteristical regions and extracted the interested targets。Image segmentation as a technical prerequisite for information extraction, it is a key step from image processing to image analysis。Image segmentation is of great significance to promote the application and development of remote sensing technologies。In remote sensing images, roads show strip-shaped features, it may be further extracted through separating road goals from other regions。Therefore, this paper use technologies of remote sensing image segmentation and classification to extract road features, the study focuses on the segmentation algorithms of remote sensing images。BY systemic analyzing segmentation methods of remote sensing image, this paper proposes three algorithms of image segmentation:Markov random field model, K-means clustering, and hybrid model composed by the support vector machine (SVM) classification and fuzzy C-means (FCM) clustering (this mixture model algorithm is unique in this paper), and researches and improves segmentation algorithms of image to suit to the conditions of road extraction, then achieves to extract the roads through mathematical morphology; finally, the extraction effect of three road segmentation algorithms are compared by visual and quantitative analysis。Accuracy evaluation showed that the accuracy of road extraction by the hybrid model is more than the K-means clustering and Markov random f ield。Road extraction accuracy by the hybrid model is 94.57%, K-means clustering is 82.18%, Markov random field is 85.43%。Main purpose of this paper is how to achieve recognition and extraction of the road in high-resolution remote sensing image。In the study, this paper uses a variety of segmentation algorithms to segment road informations, and main contents are:1,This paper conducts a pre-operation before segmentation of remote sensing images, the quality of pre-operation will directly affect the segmentation, especially in image enhancement。Therefore,this paper summarizes some common methods of image enhancement, and apply to pretreatment in remote sensing image;2,This paper focuses on the three segmentation algorithms:Markov random field model, K-means clustering, and hybrid model composed by SVM classification and FCM clustering, and researches and improves segmentation algorithms of image to suit to the conditions of road extraction3,After get better segmentation results, this paper uses mathematical morphology to remove all kinds of noises and achieve road information extraction; finally, the extraction effect of three road segmentation algorithms are compared by visual and quantitative analysis。...