A commonly used tool for estimating the parameters of a mixture model is the Expectation–Maximization (EM) algorithm, which is an iterative procedure that can serve as a maximum-likelihood estimator. The EM algorithm has well-documented drawbacks, such as the need for good initial values and the possibility of being trapped in local optima. Nevertheless, because of its appealing properties, EM plays an important role in estimating the parameters of mixture models. To overcome these initialization problems with EM, in this paper, we propose the Rough-Enhanced-Bayes mixture estimation (REBMIX) algorithm as a more effective initialization algorithm. Three different strategies are derived for dealing with the unknown number of components in the mixture model. These strategies are thoroughly tested on artificial datasets, density–estimation datasets and image–segmentation problems and compared with state-of-the-art initialization methods for the EM. Our proposal shows promising results in terms of clustering and density-estimation performance as well as in terms of computational efficiency. All the improvements are implemented in the rebmix R package.

Condition monitoring and fault detection are nowadays popular topic. Different loads, enviroments etc. affect the components and systems differently and can induce the fault and faulty behaviour. Most of the approaches for the fault detection rely on the use of the good classification method. Gaussian mixture model based classification are stable and versatile methods which can be applied to a wide range of classification tasks. The main task is the estimation of the parameters in the Gaussian mixture model. Those can be estimated with various techniques. Therefore, the Gaussian mixture model based classification have different variants which can vary in performance. To test the performance of the Gaussian mixture model based classification variants and general usefulness of the Gaussian mixture model based classification for the fault detection, we have opted to use the bearing fault classification problem. Additionally, comparisons with other widely used non-parametric classification methods are made, such as support vector machines and neural networks. The performance of each classification method is evaluated by multiple repeated k-fold cross validation. From the results obtained, Gaussian mixture model based classification methods are shown to be competitive and efficient methods and usable in the field of fault detection and condition monitoring. Highlights• Gaussian-mixture-model-based classification was applied to the bearing-fault classification. • To discriminate the faulty from non-faulty bearings only simple statistics from vibrational data was used. • Two different datasets, the Case Western Rice University dataset and Bearing vibration data collected under time-varying rotational speed conditions dataset are used. • The Gaussian-mixture-model-based classification method showed to be a competitive and efficient method.

A maximum-likelihood estimation of a multivariate mixture model’s parameters is a difficult problem. One approach is to combine the REBMIX and EM algorithms. However, the REBMIX algorithm requires the use of histogram estimation, which is the most rudimentary approach to an empirical density estimation and has many drawbacks. Nevertheless, because of its simplicity, it is still one of the most commonly used techniques. The main problem is to estimate the optimum histogram-bin width, which is usually set by the number of non-overlapping, regularly spaced bins. For univariate problems it is usually denoted by an integer value; i.e., the number of bins. However, for multivariate problems, in order to obtain a histogram estimation, a regular grid must be formed. Thus, to obtain the optimum histogram estimation, an integer-optimization problem must be solved. The aim is therefore the estimation of optimum histogram binning, alone and in application to the mixture model parameter estimation with the REBMIX&EM strategy. As an estimator, the Knuth rule was used. For the optimization algorithm, a derivative based on the coordinate-descent optimization was composed. These proposals yielded promising results. The optimization algorithm was efficient and the results were accurate. When applied to the multivariate, Gaussian-mixture-model parameter estimation, the results were competitive. All the improvements were implemented in the rebmix R package.

No abstract

Unsupervised image segmentation is one of the most important and fundamental tasks in many computer vision systems. Mixture model is a compelling framework for unsupervised image segmentation. A segmented image is obtained by clustering the pixel color values of the image with an estimated mixture model. Problems arise when the selected optimal mixture model contains a large number of mixture components. Then, multiple components of the estimated mixture model are better suited to describe individual segments of the image. We investigate methods for merging the components of the mixture model and their usefulness for unsupervised image segmentation. We define a simple heuristic for optimal segmentation with merging of the components of the mixture model. The experiments were performed with gray-scale and color images. The reported results and the performed comparisons with popular clustering approaches show clear benefits of merging components of the mixture model for unsupervised image segmentation.