The Study Of Information Extraction Technology For Remote Sensing Based On Feature Knowledge

With the continuously improvement of remote sensing data in spatial, spectral and time resolution, a lot of data in various types of remote sensing applications is provided.However, the abundant information obtained in remote sensing data has not been fully explored and used, which leads to a huge waste in remote sensing resource, and hinders its further application as well. Therefore, the development of remote sensing information extraction and recognition has become an urgent requirement.The extraction of remote sensing information is to extract the useful information (such as buildings, land-use types, vegetation,temperature, and other interested target) from remote sensing image data. The object of the study is the geographical entity and the related geo-phenomenon. As the entity object of remote sensing study-the information of surface space is multi-dimensional and infinite, but the remote sensing data obtained through transmission of information is a two-dimensional and simplified information, so the asymmetry between surface information and the remote sensing data makes the remote sensing information extraction(the process of geo-spatial analysis and inversion) fuzzy and get multiple solutions. To achieve the accurate extraction of the automation of remote sensing information extraction, the remote sensing data must be made full use of,meanwhile, flexibly reuse the related feature rule which change with the data source and application condition through the excavating and adding other features to form a complete description.In the stage which remote sensing information extraction can be realized by visual interpretation, the interpretation experts, through the comprehensive utilization of direct interpretation including color, shape, size, shadow, texture, pattern, location, etc, and other indirect interpretation signs including distribution and topological spatial relations, and combined with other comprehensive analysis and logical reasoning of non-remote sensing data resources, so as to achieve a higher precision of information extraction. However, this method is labor intensive and time-consuming, and rely mainly on the knowledge of experts which is difficult to reuse. The existing automatic extraction methods always bring rules and knowledge to the extraction of information by adding features of the knowledge dimension on the basis of mathematical statistics model, or through the use of neural networks, decision tree methods. Although in some applications, such methods achieved good results, it is hard to promote, mainly for two reasons:I)The feature expression often uses the pixel as the object, for the shape and characteristics of semantic integrity is difficult to express; 2) The feature knowledge rules always use a priori knowledge and invisible expression in the recognition process, which is not in conformity with the habit of human interpretation, and hard to reuse. Moreover, it is impossible to achieve self-adaptive adjustment for the sensors and imaging conditions change.Object-oriented image processing method is to analyze objects from the pixel-level to the object level, in order to get the shape, spatial relations and other features outside the spectrum, and through taking advantage of the relationship between objects, it can express a higher level of semantic feature and provide better information vector compared to pixel-level image analysis. This thesis analyzes characteristics of image objects and knowledge of the rules of expression based on object-oriented image analysis model, and establish a reusability feature knowledge base to achieve management and updates of the target feature in geographical entities, and on this basis it analyze the application mode of feature knowledge base in imaging feature building and target recognition of an object, and study key issues such as rules of optimal combination features in target recognition based on samples to achieve the extraction framework of feature-based knowledge of remote sensing information. The main content of this thesis include:(1)The framework of feature-based knowledge for Remote Sensing Information Extraction:It analyzes geological features of knowledge in the role of remote sensing information extraction, provides the framework of feature-based knowledge for Remote Sensing Information Extraction, and studies application model based on the framework for the application of information extraction.(2) The physical characteristics of geographic knowledge management and update mechanism:It analyzes the use of remote sensing information extraction of features, rules and knowledge, and numerical expression, designs the dynamic hierarchical indexing strategy to achieve knowledge management of feature knowledge.(3) The object construction technology research under the control of feature knowledge:It analyzes the effect of parameters to its object in the multi-scale image segmentation method, and builds the evaluation criteria of object establishment based on feature knowledge. According to the optimization of parameters in the process of the realization of that criterion, it finally achieves the automatic construction of the image.(4) The study of preferred method in features recognition of geographical entities:It studies the recognition method for geographical entities based on image object, establishes the way to study geographical entities automatically derived from the training set, focuses on the application requirements for updating the feature rule in feature base, studies on the way and optimized strategy of feature combination based on image acquisition methods, and finally achieves the feedback on the characteristics of knowledge base updates.This paper is divided as follows on the basis of the content of study.1)At the beginning of this chapter, it elaborates on the significance of the study of remote sensing information extraction and indicates the imperative demand of raising the level of remote sensing thematic information extraction and cognition, then point out that the knowledge-based remote sensing information extraction method is a trend that is all but irreversible; and provides fundamental basis for intelligent remote sensing information extraction on the basis of object-oriented methods. Finally, it demonstrates the content of the study and the structural arrangement of this article.2)This paper gives description of the feature-based library of remote sensing information extraction technology framework. In the first place, it introduces the geographic entity characteristics and imaging findings by narrating the remote sensing data acquisition process and analysis the function of characteristic knowledge which is used in information extraction; and then summarize the application model.Base on this, it lays out a frame of feature-based knowledge of remote sensing information extraction and analyzes the key issues.3)Construction system of the knowledge base of image feature. Firstly, it introduces the key factors of knowledge base, characteristic knowledge base and how to realize the characteristic knowledge base, and then analyzes characteristics, rules and knowledge that is used in Remote Sensing Information Extraction. Base on this, the way of storing characteristics and knowledge is given below, and the appropriate index method which fits the information extraction is provided.4)Describe knowledge-constrained approach to build an object. This chapter first introduces the multi-scale segmentation method of video object, and highlights the FENA segmentation method. Then analyze the parameters which affect the construction of objects through experiments on the scale, shape factor, bandwidth, weight, etc. Through the introduction of standards as well as genetic algorithms, it solves the automatic determination method of relevant parameters to build the object, and establish an automatic segmentation process by introducing the sample evaluation criteria. Finally this chapter accesses the new segmentation method using the experimental data.5)Give the optimization mechanism of feature combination and surface features recognition. Object-based geographical entity is recognized based on various types of features through a combination of implementation. This chapter puts forward the method of using statistical classification and fuzzy classification and multiple classifiers to further improve the classification accuracy based on the proposed use of statistical classification and fuzzy classification method 6)The paper provides the of the extraction methods based on samples. This chapter focuses on the application requirements for automatic extraction based on the rules of samples. The spectral image object, shape, semantic and other characteristics, and association rules extraction methods are introduced, and on this basis, the feature combinatorial optimization methods are proposed.7)In summarizing the main contents of the paper, it proposed the problems and inadequacies of the thesis, and focus and direction for further research as well.This thesis brings the feature knowledge to remote sensing information extraction process through the in-depth study of object-oriented model of information extraction. It solves some key issues such as management and updating mechanism of geographical entities, automatic construction of target feature knowledge-driven, recognition of the target feature combination and optimization method. The basis for rapid extraction of intelligent remote sensing is provided, and can be widely used in land resource surveys, land monitoring, basic geographic information updates, and protection of environmental resources and so on. Enhancing the spatial information in related fields can provide more accurate baseline data for analyzing and making decisions, which has nice prospects and a wide range of application fields.