Data Information Pattern Recognition Theory And Its Applications

Based on domestic and overseas related research results of statistical analysis theory, pattern recognition theory and information theory, and supported by National Natural Science Foundation project Information Pattern Recognition Theory And Its Application in Geosciences (NO. 40074001), this paper systematically studies the theories and methods of data compression and feature extraction, and makes thorough research into some problems on information pattern recognition.In view of the characteristics of surveying data, the paper raises the definition of data and its features from the view of information science. And it blazes a trail for subsequent research.Based on traditional statistic analysis theories, this paper summarizes and analyses data compression and feature extraction methods. There are three rules of data class separability: Euclidean distance rule, probability distance rule and entropy function rule. Their respective characteristics, their applied conditions and their feature extraction methods are studied in the paper. Under the rule of mean variance being the minimum, the paper studies data feature analysis methods extracting variance information with collectivity entropy and methods extracting effective classification information from class mean vectors. Based on K-L transform and principal component analysis, this paper also studies data compression and feature extraction, raises entropy function of principal components, and studies their information features from the view of information theory.This paper also summarizes and analyses two explored data analysis methods based on high-ordered statistical features: projection pursuit and independent component analysis, which include properties and characters of projection pursuit index and object function of independent component analysis, relations between negentropy and mutual information, and the approach to negentropy. Based on minimum mutual information rule, it studies maximum information accelerated algorithm of independent component analysis, and gives some experiments to verify the effectivity of this algorithm.The paper preliminarily studies feature extraction methods based on mutual information, raises its extracting rule; it also introduces Renyi entropy and the calculation of Renyi quadratic entropy and studies Renyi quadratic mutual information feature extraction methods. Judging from the above, feature extraction theory based on information maximization is a generalization of all theories and methods about feature extraction.Based on pattern recognition theory and information theory, this paper studies surveying data parent mean value's alteration by pattern recognition. And it also studies the relationship between entropy and surveying uncertainty, and presents a novel method to detect the outliers by information entropy. This paper also studies analytical model of maximum entropy, minimum description length and their application in surveying fields.Some of the research results have been published in science papers.