Expectation Maximization Model Research Articles

A Persistent Scatterer Point Selection Method for Deformation Monitoring of Under-Construction Cross-Sea Bridges Using Statistical Theory and GMM-EM Algorithm

Using traditional algorithms to identify persistent scatterer (PS) points is challenging during bridge construction because of short-term changes at construction sites, such as earthworks, as well as the erection and dismantling of temporary structures. To address this issue, this study proposes a PS point selection method based on statistical theory and Gaussian Mixture Model-Expectation Maximization (GMM-EM) algorithm. This method adopts amplitude information as an incoherence evaluation indicator. Furthermore, the statistical median of the amplitude dispersion index and amplitude mean is screened twice to extract a set of candidate points, including PS points that exhibit stable backscattering over long durations. Temporal coherence is simultaneously used as the coherence evaluation indicator. Another candidate point set is obtained by extracting high-coherence PS points using the GMM-EM algorithm. These sets of candidate points are then combined to obtain a final PS points set. In the experiment, the deformation monitoring of the under-construction Shenzhen-Zhongshan Cross-Sea Bridge in China was selected as a case study, with 28 Sentinel-1A images used as the data source for PS selection and deformation information extraction. The results show that the proposed method enhanced the density and quality of PS points on the under-construction cross-sea bridge compared to existing PS selection methods, thus offering higher reliability. Deformation analysis further revealed fluctuating deformation trends at characteristic points of the Shenzhen-Zhongshan Cross-Sea Bridge, indicating the occurrence of elastic deformation during its construction.

Online Public Opinion Analysis Model Based on Long Short-term Memory Network and Expectation Maximization

Abstract With the rapid dissemination of online public opinion, its emotions are easily transmitted to the general public. It possesses a certain level of social mobilization capacity and can impact the stability of society. To characterize the sentiment trend of social network information and determine its influence, we have proposed a method based on long short-term memory (LSTM) networks and expectation maximization(EM). This model employs a long short-term memory network for data training, obtaining the number of positive public opinions through word-to-word matching. Based on the expectation-maximization method and Jensen’s inequality, the lower bound of the maximum likelihood function is iteratively computed, ultimately achieving convergence of this likelihood function. This convergence value is then used for sentiment analysis. Our study utilizes 10,000 valid pieces of data from the Sina microblog as experimental data. By comparing our model with the K-MEANS model and the EM model, the results indicate significant improvements in the accuracy and convergence of our model. Our research discovers that the influence of public opinion increases as the compensation value for adoption rises, and the probability of public opinion generation gradually increases with the length of user registration years, eventually slowing down.

Comparison of Three Methods for Distinguishing Glacier Zones Using Satellite SAR Data

Changes in glacier zones (e.g., firn, superimposed ice, ice) are good indicators of glacier response to climate change. There are few studies of glacier zone detection by SAR that are focused on more than one ice body and validated by terrestrial data. This study is unique in terms of the dataset collected—four C- and L-band quad-pol satellite SAR images, Ground Penetrating Radar data, shallow glacier cores—and the number of land ice bodies analyzed, namely, three tidewater glaciers in Svalbard and one ice cap in Iceland. The main aim is to assess how well popular methods of SAR analysis perform in distinguishing glacier zones, regardless of factors such as the morphologic differences of the ice bodies, or differences in SAR data. We test and validate three methods of glacier zone detection: (1) Gaussian Mixture Model–Expectation Maximization (GMM-EM) clustering of dual-pol backscattering coefficient (sigma0); (2) GMM-EM of quad-pol Pauli decomposition; and (3) quad-pol H/α Wishart segmentation. The main findings are that the unsupervised classification of both sigma0 and Pauli decomposition are promising methods for distinguishing glacier zones. The former performs better at detecting the firn zone on SAR images, and the latter in the superimposed ice zone. Additionally, C-band SAR data perform better than L-band at detecting firn, but the latter can potentially separate crevasses via the classification of sigma0 or Pauli decomposition. H/α Wishart segmentation resulted in inconsistent results across the tested cases and did not detect crevasses on L-band SAR data.

An Elliptical Modeling Supported System for Human Action Deep Recognition Over Aerial Surveillance

The advancement of computer vision technology has led to the development of sophisticated algorithms capable of accurately recognizing human actions from red-green-blue videos recorded by drone cameras. Hence, possessing an exceptional potential, human action recognition also faces many challenges including, tendency of humans to perform the same action in different ways, limited camera angles, and field of view. In this research article, a system has been proposed to tackle the forementioned challenges by using red-green-blue videos as input while the videos were recorded by drone cameras. First of all, the video was split into its constituent frames and then gamma correction was applied on each frame to obtain an optimized version of the image. Then the Felzenszwalb’s algorithm performed the segmentation to segment out human from the input image and human silhouette was generated. Utilizing the silhouette, skeleton was extracted to spot thirteen body key points. The key points were then used to perform elliptical modeling to estimate the individual boundaries of the body parts while the elliptical modeling was governed by the Gaussian mixture model-expectation maximization algorithm. The elliptical models of the body parts were utilized to spot fiducial points that if tracked, could provide very useful information about the performed action. Some other features that were extracted for this study include, the 3d point cloud feature vector, relative distance and velocity of the key-points, and their mutual angles. The features were then forwarded for optimization under a quadratic discriminant analysis and finally, a convolutional neural network was trained to perform the action classification. Three benchmark datasets including, the Drone-Action dataset, the UAV-Human dataset, and the Okutama-Action dataset were used for a comprehensive experimentation. The system outperformed the state-of-the-art approaches by securing accuracies of 80.03%, 48.60%, and 78.01% over the Drone-Action dataset, the UAV-Human dataset, and the Okutama-Action dataset respectively.

Analysis of Monaural and Binaural Statistical Properties for the Estimation of Distance of a Target Speaker

The paper presents an auditory distance perception model that is based on the extraction of statistical properties from monaural and binaural features in a reverberant room environment. The developed framework has considered both mono and stereo speech signals originated from different distances at various reverberation time periods. Hence, two models, namely single-channel monaural statistics and binaural-channel monaural statistics, have been discussed in this study. The distance-dependent statistical features from fused monaural coefficients, namely cepstral and envelope features, are chosen as an input to the different classification algorithms such as Gaussian mixture model-expectation maximization, support vector machine and random forest for the estimation of distance of a desired target user. The monaural coefficients are extracted in addition with the binaural cues, such as interaural time and level differences and interaural coherence (ITD, ILD and IC) for the binaural speech signals and eventually applied for the estimation of distance. The proposed monaural and binaural models observe an average of more than 5% better results compared to existing baseline techniques even at lower signal-to-noise ratio, 0 dB.

Detecting Clusters in Atom Probe Data with Gaussian Mixture Models.

Accurately identifying and extracting clusters from atom probe tomography (APT) reconstructions is extremely challenging, yet critical to many applications. Currently, the most prevalent approach to detect clusters is the maximum separation method, a heuristic that relies heavily upon parameters manually chosen by the user. In this work, a new clustering algorithm, Gaussian mixture model Expectation Maximization Algorithm (GEMA), was developed. GEMA utilizes a Gaussian mixture model to probabilistically distinguish clusters from random fluctuations in the matrix. This machine learning approach maximizes the data likelihood via expectation maximization: given atomic positions, the algorithm learns the position, size, and width of each cluster. A key advantage of GEMA is that atoms are probabilistically assigned to clusters, thus reflecting scientifically meaningful uncertainty regarding atoms located near precipitate/matrix interfaces. GEMA outperforms the maximum separation method in cluster detection accuracy when applied to several realistically simulated data sets. Lastly, GEMA was successfully applied to real APT data.

A Study of Gearbox Residual Life Prediction Method Based on Stochastic Filtering and Recursive Expectation Maximization Model

Residual life prediction is a critical and difficult problem in condition-based maintenance decision-making. Aiming to deal with the problems that practical data is limited and the estimation of initial parameters is not accurate in maintenance practice, a residual life prediction method for gearbox based on stochastic filtering (SF) is proposed. In this method, recursive expectation maximization (REM) algorithm is introduced to update the parameters, and a maximum likelihood estimation method is designed to update the unknown parameters. Finally, the validity and practicability of the model are validated by an example.

LabelBoost: An Ensemble Model for Ground Truth Inference Using Boosted Trees

We introduce LabelBoost, an ensemble model that utilizes various label aggregation algorithms to build a higher precision algorithm. We compare this algorithm with majority vote, GLAD and an Expectation Maximization model on a publicly available dataset. The results suggest that by building an ensemble model, one can achieve higher precision value for aggregating crowd-sourced labels for an item. These higher values are shown to be statistically significant.

Improved Generative Semisupervised Learning Based on Finely Grained Component-Conditional Class Labeling

We introduce new inductive, generative semisupervised mixtures with more finely grained class label generation mechanisms than in previous work. Our models combine advantages of semisupervised mixtures, which achieve label extrapolation over a component, and nearest-neighbor (NN)/nearest-prototype (NP) classification, which achieve accurate classification in the vicinity of labeled samples or prototypes. For our NN-based method, we propose a novel two-stage stochastic data generation, with all samples first generated using a standard finite mixture and then all class labels generated, conditioned on the samples and their components of origin. This mechanism entails an underlying Markov random field, specific to each mixture component or cluster. We invoke the pseudo-likelihood formulation, which forms the basis for an approximate generalized expectation-maximization model learning algorithm. Our NP-based model overcomes a problem with the NN-based model that manifests at very low labeled fractions. Both models are advantageous when within-component class proportions are not constant over the feature space region “owned by” a component. The practicality of this scenario is borne out by experiments on UC Irvine data sets, which demonstrate significant gains in classification accuracy over previous semisupervised mixtures and also overall gains, over KNN classification. Moreover, for very small labeled fractions, our methods overall outperform supervised linear and nonlinear kernel support vector machines.

Image segmentation by EM-based adaptive pulse coupled neural networks in brain magnetic resonance imaging

We propose an automatic hybrid image segmentation model that integrates the statistical expectation maximization (EM) model and the spatial pulse coupled neural network (PCNN) for brain magnetic resonance imaging (MRI) segmentation. In addition, an adaptive mechanism is developed to fine tune the PCNN parameters. The EM model serves two functions: evaluation of the PCNN image segmentation and adaptive adjustment of the PCNN parameters for optimal segmentation. To evaluate the performance of the adaptive EM–PCNN, we use it to segment MR brain image into gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF). The performance of the adaptive EM–PCNN is compared with that of the non-adaptive EM–PCNN, EM, and Bias Corrected Fuzzy C-Means (BCFCM) algorithms. The result is four sets of boundaries for the GM and the brain parenchyma (GM + WM), the two regions of most interest in medical research and clinical applications. Each set of boundaries is compared with the golden standard to evaluate the segmentation performance. The adaptive EM–PCNN significantly outperforms the non-adaptive EM–PCNN, EM, and BCFCM algorithms in gray mater segmentation. In brain parenchyma segmentation, the adaptive EM–PCNN significantly outperforms the BCFCM only. However, the adaptive EM–PCNN is better than the non-adaptive EM–PCNN and EM on average. We conclude that of the three approaches, the adaptive EM–PCNN yields the best results for gray matter and brain parenchyma segmentation.

Blind source separation of speech sources in wavelet packet domains using Laplacianmixture model expectation maximization estimation in overcomplete cases

Speech processing has benefited a great deal from wavelet transforms. Wavelet packetsdecompose signals into broader components using linear spectral bisecting. Themixing matrix is the key issue in the blind source separation literature especially inunderdetermined cases (more sources than sensors). In this paper, algorithms are proposedfor estimating the mixing matrix and separation of speech signals from noise free linearmixtures in overcomplete cases. Mixtures of speech signals are decomposed using waveletpackets, the phase difference between the two mixtures is defined and used in the waveletdomain, and histograms of phase differences are obtained for every wavelet packet. In ourmethod, the Laplacian mixture model is considered in the wavelet packet domain, and isapplied to each histogram of packets. An expectation maximization algorithm is used totrain the model and calculate the model parameters. We also propose a novelmethod for obtaining the best wavelet packet node for finding source directions inscatter plots using variance calculations. A comparison is made to evaluate theperformances of different mother wavelets in the estimation of the mixing matrixand the best wavelet has been chosen. On the basis of the geometrical model, atwo-step adaptive algorithm is proposed for separating sources from mixtures.

Interacting multiple-model-based tracking of multiple point targets using an expectation maximization algorithm in an infrared image sequence

Data association and model selection are important factors for tracking multiple targets in a dense clutter environment without using a priori information about the target dynamic. We propose an interacting multiple model-expectation maximization (IMM-EM) algorithm by incorporating different dynamic models for the target and using Markov random field (MRF) for data association. In this way we are able to track maneuvering and nonmaneuvering targets simultaneously in a single batch mode (sequential). Moreover, it can be used for real-time application. The proposed method overcomes the problem of data association by incorporating all validated measurements together using an EM algorithm and exploiting MRF. It treats the data association problem as an incomplete data problem and measurement association as missing data. In the proposed method, all validated measurements are used to update the target state, and the probability density function (pdf) of observed data, given a target state and measurement association, is treated as a mixture pdf. This allows us to combine the likelihood of a measurement due to each model. The association process now incorporates IMM, and consequently, it is possible to track any arbitrary trajectory. We also consider two different cases for association of measurement to target: association of each measurement to the target is independent of each other; and association of a measurement influences an association of its neighbor measurement.

Image identification and restoration in the subband domain

When faced with a large support point spread function (PSF), the iterative expectation-maximization (EM) algorithm, which is often used for PSF identification, is very sensitive to the initial PSF estimate. To deal with this problem, the authors propose to do EM image identification and restoration in the subband domain. After the image is first divided into subbands, the EM algorithm is applied to each subband separately. Since the PSF can be taken to have smaller support in each subband, these subbands should be less of a problem with the EM model identification. They also introduce an adaptive subband EM method for use in the upper frequency subbands.