Study And Implemention Of GIS Oriented Remote Sensing Imagary Interpretation

Geographic Information System (GIS) is one of the computer systems designed to collect, store, query, analyze and display geographically referenced data. Geo-referenced data, also known as geospatial data is used to describe the location and spatial attributes of data elements, such as roads, parcels, buildings and vegetation cover. GIS is outstanding capable of processing and analyzing geo-referenced data. Remote sensing imagery data is one of the most important components of geographic information system database. With the development of remote sensing technology, human-beings have acquired vast amounts of remote sensing data. At present, most remote sensing images are processed by artificial means, spending a lot of manpower and resources. Today, with the rapid development of computer hardware technology, the computing speed and processing power for the interpretation of vast amounts of remote sensing image data provides enough hardware foundation.Interpretation of remote sensing images means computer simulation of the human brain's way of thinking, using artificial intelligence to complete the decomposition and transformation of the image and, ultimately, obtain GIS vector data and raster data. When human brain tends to decompose a regional scale corresponding to the geometry of raster GIS data structure, necessary methods are required. At first, extract the relevant parts of the image. Then classify them according to common sense and experience, and add result to the GIS coordinate system. During road or linear model extraction, human brain tends to randomly find a starting point for tracking, access to the linear trend model, and make result in GIS vector data. Thus, remote sensing image interpretation system could be designed and developed. The system should be able to interpret remote sensing image into GIS data structure for docking, reduce manual intervention as much as possible, and result in certain context accuracy.In order to balance between the computational cost and precision request of image segmentation, an approach aims at parallel implementation of an improved region growing solution applicable and adaptable to multiple matters and data types is presented. Each image analysis problem deals with structures of a certain spatial scale and starts with corresponding resolution value. This algorithm is achieved by a general segmentation based on homogeneity definitions in combination with local and global optimization techniques with automatic custom input resolutions. Considerably, with less input and more parallel computational capabilities, better time complexity is provided with acceptable loss of significance.In this thesis, the research work and achievements include:(1) A GIS oriented HRRSIm interpretation system is designed and implemented, which consists of three modules: image segmentation, pattern recognition and geographic registration. The input of the system are remote sensing images while the output is an XML file which can dock directly with the GIS data structure;(2) A two-step image segmentation algorithm is presented to meet the needs of GIS. They are scalar segmentation and vector segmentation. Scalar segmentation is an object-oriented multi-scale image segmentation method, using multi-scale technology to avoid over-segmentation phenomenon. Vector space segmentation is an algorithm of the Quadratic snake model, acquiring anti-noise ability and is able to get precise vector of information;(3) During scalar segmentation stage, a dynamic two-dimensional Gaussian distribution function weighted spectral heterogeneity algorithm is proposed. It weakens the edge effect influence of the objects and achieves more accurate segmentation results;(4) During pattern recognition stage, fuzzy features of objects are proposed as the input of the ambiguity function. Scalable library of fuzzy functions and fuzzy rules is achieved and expert data can be added and reused;(5) At the end of the pipeline in the system, geo-reference is proposed to solve two-step registration process to obtain the synchronization of data and improve the interface between vector and raster data integration;(6) Hardware acceleration approaches are studied and implemented in image segmentation module. With the implementation of multi-threading and parallel computation methods, the system takes full advantage of existing computer hardware capabilities, and the calculation speed of the module upgrading to nearly 3 times.