Research Of Intelligent Information Extraction Of Marine Images

Along with the development of science and technology, the relationship between human and marine has become more and more closely. Marine not only provides necessary material and energy for human beings to support the survival and development, but also plays an important role in climate change and ecology. The research of marine environmental monitoring and marine resources exploration need marine information to serve as a foundation, and the marine information technology has become a new developing field in recent years. For several decades, marine satellite remote sensing technology has accumulated massive, dynamic, multi source and multidimensional marine environmental data for marine monitoring and investigation, however, the research of information extraction technology is lagging behind data acquisition, valuable data resources potential can not be fully developed and utilized, and people are now facing the problem of "rich data and poor knowledge". The theory and technology study of information from marine satellite images is one key to exploit and protect marine.Intelligent information extraction is to apply computational intelligence methods such as: knowledge discovery, machine learning, to process domain data. It is a kind of explorative data analysis to discover previously unknown and potentially useful knowledge from massive, complicated and diversified dataset. In the past twenty years, its theory research and technical application have been increasingly enriched and it has provided human a powerful tool to know the world.This dissertation takes marine satellite image information intelligent extraction as the research object, it combines computational intelligence methods, image information structure and marine satellite remote sensing together. An abstract marine satellite image information model has been introduced and concrete intelligent methods have been studied, finally a web based marine satellite image information platform has been designed and implemented. The main contents and innovations are summarized as follows.1. Application Innovation:Manifold learning based sea surface temperature image analysis. Nonlinear dimension reduction is employed to get the low dimensional manifold structure from high dimensional sea surface temperature images, that is, to find the embedded projection. The dissertation discusses the relation between intrinsic space dimensionality and space time range. With the help of dimension meaning, the characteristics of low dimensional space such as projection distribution, anomaly point, have been researched. The result of experiment shows that the exception distance corresponds to ONI index and it has certain ability to judge El Nino phenomenon.2. Procedure Innovation:Kernel method based marine mesoscale eddies recognition. In this dissertation, kernel method is employed to measure the similarity of eddies, and a structural statistical function is described to predict the location of eddy center. Sliding window algorithm is proposed to detect multiple eddies in one sea current image. Oceanography knowledge (feature of mesoscale eddy) and SSE technique is adopted to accelerate the speed of eddy classification. The experiment shows that this method can get higher accuracy (97.6%) of eddy recognition than current vector constraint algorithm.3. Technology Innovation:Design and implementation of web based operational marine satellite image information platform. After analyzing the pattern of marine satellite image information (original and product) management and sharing in web environment, an operational workflow of marine scientific research is built. Service-Oriented Architecture (SOA) is taken to design and implement a web-based distributed marine satellite information platform. The platform has already been used in several projects.The research result indicates that the abstract marine satellite image information extraction model of this dissertation can discover oceanography knowledge from different perspectives. The field based image analysis retrieves low dimensional special knowledge of sea surface temperature field and the object based method analyzes the distribution feature of mesoscale eddies through supervised algorithm. The marine satellite information platform can provide user centered information service under a network environment.