Multi-Source Information Feature Extraction And Fusion And Its Application In Information Management

Numerous Information, which has massive, multi-source, high-dimensional, nonlinear and dynamic characteristics, is emerging in the application of sensor technology, and this kind of information makes information feature extraction and fusion more complicated. At present, the theories and methods of multi-source information feature extraction and fusion have been widely applied in the field of scientific research, economy, military, and so on. The theories and methods of multi-source information feature extraction and fusion have gradually formed their own research system, which is built on the basis of multi-source information feature extraction, with core of multi-source information fusion, and with the application means of oriented multi-source information intelligent system. However, how to extract the feature of multi-source images information effectively, and then how to fuse this kind of information are still the hot topics and problems with the attention from the scholars in our country and overseas. Therefore, the research on multi-source information feature extraction and fusion is theoretically meaningful and practically valuable to strengthen the ability of image information management under the complicated environment.For the nonlinear, high-dimensional and uncertain characteristics of multi-source information, the dissertation has studied its feature extraction and fusion and its application in the field of information management. So, the dissertation has not only extended the classical theory and method of information feature extraction and fusion, but also greatly improved the adaption ability and application scopes of information feature extraction and fusion in the industry and daily life. The details are as follows:(1) According to the multi-source, high-dimensional and nonlinear characteristics of the information caught by sensors, the dissertation has analyzed the problems of locally linear embedding and fractal dimension estimation. Then, genetic algorithm and isometric mapping have been used respectively to improve the above two methods. Next, the dissertation has suggested two new methods, they are locally linear embedding method based on genetic algorithm and fractal dimension estimation method based on isometric mapping. The performance of the above two new methods has been compared and analyzed through several contrast experiments.(2) According to the obtained multi-source information often with the characteristics as being massive, high-dimensional and unlabeled, the dissertation has analyzed the problems such as the influence of similarity measure of cluster on cluster effect and high-dimensional and complex data clustering during the partitioning process of information. On the basis of above studies, Spectral Clustering Locally Linear Embedding, Fast Global K-means based on Manifold Distance and Dimension Reduction and Multi-Phase Clustering based on Manifold Distance and Dimension Reduction have been suggested respectively. The reliability and accuracy of the above three new methods have been examined through several contrast experiments.(3) According to the uncertainty of the user-facing information environment, the dissertation has analyzed the problems such as the irrational assignment of conflict evidence, bad convergence effect of fusion method and multi-BBM evidence reasoning during the combination process of D-S evidence theory. On the basis of above studies, the concepts of similarity of low-dimensional evidence bodies and multi-source information fusion method based on D-S evidence theory have been proposed. The ability to solve above-mentioned three problems and the practical applying value of the new method have been examined through conflict evidence contrast experiment, multi-BBM contrast experiment, simulation information experiment and real information experiment.(4) The dissertation has researched on the application of multi-source information feature extraction and fusion in the field of information management. For the artificial recognition methods are difficult to analyze and recognize the multi-source and massive image information in image information management, these methods proposed in the previous chapters has been used as multi-source information feature extraction and fusion methods in intelligent image information recognition systems, the contradictions between multi-source and massive image information, single analysis method and limited processing resources has been resolved better, and the corresponding intelligent image information recognition strategies has also been established. Then, the architecture of intelligent image information recognition system has been built based on the analysis of the tasks and functions in intelligent image information recognition system using object-oriented modeling methods. On the basis of above studies, the dissertation has also implemented a prototype system after planning hierarchical structures and designing function module for intelligent image information recognition system under Matlab development environment.