Research And Implementation Of Fuzzy Clustering Algorithm

Fuzzy clustering is one of the important means for effectively dividing a data set without labels.With the arrival of the era of big data,the amount of data has grown exponentially.However,most of the data are not labeled.How to accurately classify these data and provide more accurate services to users has become the focus of research in the society today.The classical fuzzy clustering algorithm FCM is widely used because of its simplicity and high efficiency,but its membership degree is 1.The noise points and outliers in the dataset have a great influence on the final clustering result.? The PCM algorithm breaks the limit of the membership degree and 1 and reduces the influence of noise points and outliers.When FCM and PCM algorithms are clustering high-dimensional data sets,the computational efficiency is often very low.The introduction of kernel functions greatly improves the efficiency of the algorithm for calculating high-dimensional data sets.KFCM and KPCM is proposed on the basis of FCM and PCM respectively.However,the above four clustering algorithms still have two major problems: ignoring the relationships between the elements of the classes and the initial cluster centers are randomly selected.This article studies the relationship between the elements of the interclass and the initial clustering center,and obtains the following results:In view of the traditional clustering algorithm which only considers the relationship between the elements in the class to ignore the relationship between the class and the class,when the data set with ambiguous boundary is processed,the problem of misclassification of the boundary point will be caused.Based on KPCM,Based on Improved Kernel Possibility C-means Maximization Clustering Algorithm.The algorithm combined with the objective function of KPCM,and then imposed a great penalty,making the distance between the class and the center of the class widened,thus taking into account the relationship between the elements of the class,can better divide the sample at the boundary.For the problem that the initial clustering center is selected randomly,the final result of the traditional clustering algorithm is not stable.According to the compactness information of the sample distribution,the minimum clustering error is used to optimize the initial clustering center.According to the spatial distribution information of samples,the initialization algorithm calculates the sample variance by calculating the variance of the samples and selects the sample points with the smallest variance and the sample point within a certain range as the initial cluster centers to achieve an improved fuzzy clustering algorithm.Because this algorithm can obtain better initial clustering centers,it can obtain better clustering results.However,the time and space complexity of this method is high,and it is inefficient when dealing with a large number of data sets such as images.This paper studies several classical methods in image segmentation.Pixels with similar grayscale values and presence of noise points in the image will result in unsatisfactory segmentation,In response to the above problems,we tried to apply the KPMM algorithm proposed in this paper with high operating efficiency to image segmentation.Comparing experiments with FCM,PCM,KFCM and KPCM algorithms,The experimental results show that the application of KMPCM in image segmentation is better.