Research And Implementation Of Key Technologies For High Resolution Road Extraction

With the development and popularization of high-resolution remote sensing technology,the spatial resolution,temporal resolution and spectral resolution of images are getting higher and higher,and the detailed information of roads at different levels and different levels has become richer.High-resolution images have become an important data source for road information extraction and update in the construction of smart cities.In addition to road information on high-resolution images,there is a lot of non-road information such as buildings,vegetation occlusion,and moving targets on the road surface,that is,interference information.At the same time,roads show many complex features on high-resolution images.The color characteristics and widths of different types of roads are obviously different,which brings difficulties to road extraction research.Therefore,how to efficiently and accurately extract the roads on high-resolution images has been the research hotspot and difficulty of scholars at home and abroad.In view of the above problems,this paper intends to extract and optimize road results from multiple approaches such as multi-scale segmentation,seed point selection,road feature extraction and calculation,construction of SVM classifiers,mathematical morphology,and fusion of multi-result methods.The specific research work is as follows:(1)For high-resolution images that contain urban roads but have a complex background,the image segmentation scale is essential for image classification and road extraction.A single segmentation scale cannot classify multiple features well,so this article uses SLIC Superpixel segmentation and DBSCAN density-based clustering perform multi-scale segmentation,and select and sample the seed points of roads and non-roads.By calculating the characteristics of two types of seed points,such as spectral characteristics,shape characteristics and texture characteristics,to As a training sample of the SVM classifier,the road and non-road classification of the same image are divided under different segmentation scales,and the multi-scale road results that are more suitable for complex backgrounds are extracted,and the road is extracted by SVM classification.(2)For high-resolution images with relatively simple backgrounds including towns,overpass roads,elevated roads,etc.,in order to improve extraction efficiency,this article uses small sample size,grayscale processing,threshold segmentation and median filtering as preprocessing.Adopt and realize the method of automatically extracting roads based on mathematical morphology,including corrosion,dilation,closed operation and open operation in morphology,and realize the automatic extraction of township roads,overpass roads and elevated roads.And the post-processing of the road extraction of the SVM classification of the above multi-scale segmentation basically removes the non-road patches,repairs the road holes,and smoothes and optimizes the road results.(3)In order to make the results of road extraction by multi-scale segmentation and SVM classification make up for each other’s defects,this paper innovatively proposes an optimization algorithm based on Sugeno’s fuzzy integral fusion road results,which is extracted from the road extraction method based on multi-scale segmentation and SVM classification The quality weights of the road probabilistic images were merged with fuzzy integrals,and the fusion results were optimized using the mathematical morphology post-processing methods researched and implemented above.Experiments show that the algorithm of road result fusion is better than multi-scale segmentation and The road extraction results of SVM classification can extract roads more accurately,and the road result quality is higher.