Research On Object-oriented Remote Sensing Image Information Extraction Technology

With the rapid development of remote sensing information access technology, how to improve the speed and the accuracy of the remote sensing image information extraction has become an important research direction in the remote sensing data processing field.Traditional remote sensing image information extraction technology is based on pixel,and the factor involved in information extraction is the pixel's spectral information, which is often difficult to obtain satisfactory effects. In recent years,with the extensive application of high-resolution remote sensing image,the object-oriented remote sensing image information extraction technology has become widely concerned.The object-oriented technology use the"homogeneous object" which their spectrum and texture characteristics are similar as the smallest unit of information extraction, not only consider spectral information but also texture and spatial structure information of remote sensing data, and even including experience and knowledge of human Visual recognition as a basis for information extraction,So this technology can improve the accuracy and efficiency of remote sensing information extraction significantly, and then has gradually shown its unique application value and potential in remote sensing image information extraction.This paper based on the depth study of object-oriented remote sensing image information extraction techniques,uses eCognition as a tool,which is the world's first professional object-oriented classification software developed by Definiens company in Germany, and introduces object-oriented classification concept and technique into remote sensing image information extraction field. This paper uses different remote sensing image data, through multi-scale image segmentation and parameter selection,establishing class hierarchy structure and semantic structure, classification rule references, conducts a lot of tests and research.In this paper, information extraction research in three aspects has generally already done.First, using radar image to do rivers information extraction test, this test is mainly to study on how to take full advantage of the linear characteristics based on the sub object to complete the extraction of the parent object information.Second, using high-resolution aerial image to study how to conduct segmentation based on classification, how to conduct actual border optimization processing based on the results of image classification and then reach the border optimization purpose eventually. Third, use high-resolution remote sensing image data, based on image pre-processing and then execute multi-scale image segmentation, according to the spectrum and spatial features or a combination of two features, combine with the use of nearest neighbor method and membership function method, complete Huizhou City and the surrounding areas urban land use information extraction and the accuracy examination, thus it can provide an effective way with urban land use information extraction.