Urban Area Retrieval Technique Research Based On High Resolution Remote Sensing Image

With the development of space remote sensing technology,the remote sensing image data grows very rapidly. But the level of the remote sensing application is significantly lower than the level of the development of space remote sensing technology, which results in a large number of data. Thus the end users cannot extract useful information effectively, which causes huge waste of resources. It is an urgent problem to study how to rapidly and accurately find the information stored in the massive remote sensing data in the remote sensing application. Content-based remote sensing image retrieval method is the key solution to this problem, which is also the difficulty faced by massive remote sensing image retrieval in recent years.Since the 1990's, a large number of content-based image retrieval technologies have been developed, and have achieved fruitful results. However, the special study of remote sensing image retrieval cannot meet the users' requirements. Both the theoretical system and application system are still not well developed. Because the amount of remote sensing image data is huge, has wide coverage, its theme is not clear, and has multi-temporal and rich semantic characteristics, the generic image retrieval research achievements cannot be directly used in remote sensing image retrieval. For a perfect content-based remote sensing image retrieval system, the data organization, storage and management, characteristics description and extraction, similarity measure, relevant feedback, the network service model, system architecture design and implementation of theoretic framework are facing many difficulties and deficiencies, the research of key technology is imperative.Based on the comprehensive study of the developed retrieval systems and large amount of theoretical research work home and abroad, this thesis analyzes the image retrieval in the involved key technologies and research status, and comprehensively considers problems of image retrieval approaches and the end users' demand. The author also combines Bayesian Networks and machine learning methods and proposes three approaches suitable for remote sensing image retrieval. Particularly, the author takes the semantic concept "urban areas" as the key research object, deeply conducts the corresponding research, and finally proposes urban area retrieval algorithm. Considering the retrieval accuracy and time efficiency, the thesis proposes a stepwise remote sensing image retrieval algorithm, which combines semantic-based image retrieval technology and content-based image retrieval technology. The corresponding experiments are carried out based on the Matlab platform.The major research contributions of this thesis are as follows: Co-occurrence region-based Bayesian network image retrieval is proposed, which establishes the corresponding relation between image and semantic concepts through the Bayesian networks and implements the semantic-based remote sensing image retrieval; this thesis proposes a remote sensing image retrieval algorithm, which is suitable for urban area retrieval. This approach combines cooccurrence region-based Bayesian network image retrieval with the average intensity of high frequency signal, and effectively improves the semantic retrieval precision of the "urban area"; the author also proposes a new distributed remote sensing image retrieval algorithm, which significantly reduces the image retrieval time. Since the proposed algorithm considers the semantic information of images during the retrieval procedure, the precision of the prototype image retrieval system is greatly improved.