Research On Robust Fuzzy Weighted C-ordered-means Clustering Algorithm

As an unsupervised data analysis method formed by the intersection of information science,statistics,mathematics and machine learning,fuzzy clustering analysis is an important research content in the fields of data analysis,knowledge discovery and intelligent decision-making.In this paper,based on the profound understanding of the relevant research on fuzzy clustering at home and abroad,a more in-depth study on fuzzy clustering is carried out by using the combined method of theoretical analysis,data experiment and application research.This paper makes the following improvements to the existing algorithms from the following aspects: the introduction of local spatial information is used to solve the problem of the influence of neighborhood data on the data;the introduction of data typicality is used to solve the problem of the influence of data on the calculation and the distance between the centers;the introduction of the intuitionistic fuzzy algorithm is used to solve the problem of qualitative research on uncertainty of data.By solving the above-mentioned problems,the clustering algorithm is made to have better robustness.Firstly,a new hybrid clustering algorithm is proposed to solve the poor denoising,large computation and insufficient robustness of the existing fuzzy clustering methods.This new algorithm is compatible with Fuzzy C-means(FCM)clustering,clustering spatial neighborhood and typicality of clustering data.It comprehensively considers the advantages of the usage of fuzzy mathematics in dealing with the data fuziness,the spatial neighborhood objectivity of data affected by neighborhood data and the distance of the data from the center.The experimental results show that the reasonable setting of the weights of different indicators can solve the insufficient robustness and poor denoising of existing clustering algorithms,thus achieving a better clustering of a given data set.Secondly,an improve FCM algorithm called Bias-correction Fuzzy Weighted C-ordered-means clustering algorithm(BFWCOM)is proposed to solve the sensitivity of the existing FCM algorithm to noise and its insufficient consideration of data space.This algorithm uses the fuzzy local(spatial and gray)similarity measurement,and meanwhile adds the analysis of the typicality of data attributes to membership degree to ensure the insensitivity to noise and the preservation of image details.Thirdly,a new algorithm called kernel-based robust bias-correction fuzzy weighted c-ordered-means clustering is proposed to solve the large computation of BFWCOM algorithm and the weak ability of FCM clustering algorithm in processing image details.Based on kernel-induced distance measure,the algorithm combines fuzzy local(spatial and gray)similarity measurement and typicality of data attributes.In the construction of the objective function,the algorithm solves the problems of space constrained fuzzy c-means clustering algorithm,such as insensitivity to image noise and large amount of computation.The central idea of the algorithm is to map the problems that are not easy to solve in low dimension space to high dimension space by kernel function,so as to simplify the calculation and enhance the robustness of the algorithm.Finally,an intuitionistic fuzzy typicality weighted c-ordered means clustering algorithm based on intuitionistic fuzzy set theory is proposed to solve the poor performance of existing clustering algorithms in processing image data with many uncertain factors.On the basis of membership degree and non-membership degree,the concept of hesitation degree is added to the clustering algorithm,making it possible to conduct qualitative research on the uncertainty of data.In addition,this algorithm requires less calculation than the previous one,can be better applied to image denoising,and overcome the sensitivity to image noise,with stronger robustness.