Based On Fuzzy Clustering Theory Of Pattern Recognition Research

Fuzzy clustering analysis is an important branch of fuzzy pattern recognition, it is an unsupervised pattern recognition method, and is widely used in many fields. In this dissertation, the principal contents of research including: First, clustering results are compared among three methods in common use of project, namely transitive closure method of fuzzy equivalent relation and maximal tree method of fuzzy similar relation and fuzzy c-means algorithm; similarity and difference of their clustering results are analyzed from algorithms' own angle, and their utility ranges are given. Although clustering effect of fuzzy c-means algorithm is better than ones of other two methods, the principal defects and application of fuzzy c-means algorithm has been initial sensitivity and convergence speed slow and not cut out for large discrepancy of every class specimen number. And then, in view of above-mentioned defects, based on the full and systematic research of relevant theories and methods about fuzzy clustering methods, some modified fuzzy clustering methods are proposed, their principal contents include: based on being limitation of initial sensitivity about fuzzy c-means algorithm's offline learning method, an online recursive learning method has been proposed from recursive angle; based on incorporating advantages of offline learning method and online recursive learning method, a modified offline learning approach is derived from them, the approach has not only decreased initial sensitivity but also accelerated convergence of objective function, thereby contracted clustering course of fuzzy c-means algorithm; based on fuzzy c-means algorithm having limitation of equal demarcation trend for data sets, optimum clustering result of fuzzy c-means algorithm mightn't be valid demarcation for data sets of large discrepancy of every class specimen number, a little known knowledge is regard as partial supervised information, and distributing density size of data dot is regard as weighted value, a dot density weighted fuzzy c-means algorithm and a partial supervised and weighted fuzzy c-means algorithm are proposed, and calculation of weighted coefficient and choice of dot density range restriction value are objective, the simulation result proves that algorithms have not only to certain extent overcome limitation of fuzzy c-means algorithm, but also been favorable convergence, and in algorithm's utility range, clustering effects are obviously improved; based on clustering analysis having leading limitation that no matter what data sets structure is given, it is always able to classify for data sets, hence clustering validity of clustering algorithm is evaluated by clustering validity function in this dissertation; and clustering results of clustering algorithms are validated and analyzed by MATLAB language and type IRIS data sets of international recognized compare clustering result performance and other data sets. Finally, the in-depth research about applications of fuzzy clustering methods in pattern recognition is carried out: aiming at the present issue of groundwater quality evaluation in our country, fuzzy clustering method is incorporated with fuzzy comprehensive evaluation, a set of integrated methods is proposed from water area demarcation to water quality evaluation, then evaluation result of the method is compared with one of nemero index method used by our country norm, hereby rationality of the method proposed here is testified; and preliminary search is proceeded in image segmentation domain's applications with fuzzy clustering method.