Rural Road Extaction Base On Object-oriented Method With High-resolution Remote Sensing Image

Now the resolution technology of remote sensing image develops rapidly. To catch up with this trend, the image extraction technology needs to improve efficiently and accurately among the numerous image data. But, on most of situation, usually we use labor to do image extraction work. This method is of low efficiency and requires professional experience of work staff. Here, this paper aims to find way to increase the image analysis efficiency of remote sensing technology.Roads extraction takes a great part of sensing remote image extraction. According to the way of extraction, we here have two methods to do extracting:one is automatic extraction, the other one is semi-automatic extraction method to do the research of sensing remote image extraction.On the aspect of research object, we can divide the methods into Pixel-oriented method and Object-oriented method. Here needs to point out that the object-oriented method is different from the traditional one. Image will not be presented by individual pixel, but by the combined of important similar image information and related relations. Image objects are subject to a set of similar and consistent pixel among the individual data. Thanks for the more meaningful of image object, we can use the spectral characteristics, spatial texture and shape and size characteristic to do road extraction more available.After reading plentiful research papers, the writer here adopt the methods of object-oriented and pixel-oriented to analysis image of north-east part of Changshou District of Chongqing City. The remote sensing images are from CBERS02B HR data in Sep.2009. The image resolution is2.4meter.Pixel-oriented method:Firstly, writer has worked noise reduction processing and convolution processing. Secondly, by using of non-monitoring approach to do roads extraction, the writer has done morphology processing to the results. And lastly, gives the conclusion and comment for the final result.Object-oriented method:Firstly, the writer has compared and classified the different results of segmentation scale, Shape index, Compact Index after image segmentation. By using of eCognition classification features, we can build up a knowledge base for roads extraction. And lastly also gives a conclusion and comment for this method finished by using of roads growing method. Through the comparison of pixel-oriented and object-oriented methods research for roads extraction from remote sensing image. Here writer believes that object-oriented method be much more operational than pixel-oriented method during the roads extraction. We can fully use of the relations between objects or scales to build up the knowledge base. And invert the knowledge base into extraction rules, while the pixel-oriented method can only be used to do roads extraction when the master image is complete. Hence we can conclude that neither from the efficiency nor accuracy level, the object-oriented method can be more favorable to use in roads extraction research.