Parzen Window Research Articles

A soft computing automatic based in deep learning with use of fine-tuning for pulmonary segmentation in computed tomography images

The segmentation of medical images is a significant challenge for computer vision and medical techniques. In the lungs, the difficulty is mainly due to the presence of lung diseases, different dimensions of the lung, and different types and configurations of medical imaging devices. Computed tomography (CT) is a tool to aid clinical diagnosis; several systems based on computer-aided diagnostics (CAD) are developed or enhanced from CT images combined with computational methods. This study proposes an innovative approach based on the generalization of models for pulmonary segmentation in CT images, using the Convolutional Neural Network Based on Mask Regions (Mask R-CNN) combined with image processing techniques, K-means clustering, region growing and Parzen window through fine-tuning. Our models achieved satisfactory results with 97% Accuracy and 98% Dice, 99% Sensitivity, and 97% Positive Predictive Value (PPV) in our best MPK model. The method was able to generalize its learning to solve different pulmonary segmentation problems in different lung CT databases. A second set of data was used, obtaining even better results. The results obtained for this second data set were: 98% Accuracy, 99% Dice, 100% Sensitivity, and 98% PPV, demonstrating the effectiveness of our method.

Multierror stochastic gradient algorithm for identification of a Hammerstein system with random noise and its application in the modeling of a continuous stirring tank reactor

AbstractIn this article, a stochastic gradient algorithm based on the minimum Shannon entropy is proposed to identify a type of Hammerstein system with random noise. Firstly, the probability density function is estimated by a parzen window based on the Gaussian kernel. Then, the traditional stochastic gradient algorithm is adopted to estimate the parameters. However, the traditional stochastic gradient algorithm converges quite slowly. To fasten the algorithm, a multierror method is integrated into the algorithm. In this multierror gradient algorithm, the scalar error is replaced by a vector error. This vector error can accelerate the algorithm greatly and give a more accurate estimate by using the same data set. Finally, the proposed algorithm is validated by a numerical example and an industrial process. The estimation results indicate that the proposed algorithm can obtain more accurate estimates than the traditional gradient algorithm and has a faster convergence speed.

Fault Detection and Identification for Nonlinear Process Based on Inertia-Based KEPCA and a New Combined Monitoring Index

In the present study, we introduce a new approach for the nonlinear monitoring process based on kernel entropy principal component analysis (KEPCA) and the notion of inertia. KEPCA plays double roles. First, it reduces the data in the high-dimensional space. Second, it constructs the model. Before data reduction, KEPCA transforms input data into high-dimensional feature space based on a nonlinear kernel function and automatically determines the number of principal components (PCs) based on the computation of the inertia. The retained PCs express the maximum inertia entropy of data in the feature space. Then, we use the Parzen window estimator to compute the upper control limit (UCL) for inertia-based KEPCA instead of the Gaussian assumption. Our second contribution concerns a new combined index based on the monitoring indices T2 and SPE in order to simplify the detection task of the fault and prevent any confusion. The proposed approaches have been applied to process fault detection and diagnosis for the well-known benchmark Tennessee Eastman process (TE). Results were performing.

A New Method for Estimating High Frequency Attenuation Parameter of Kappa

The Fourier amplitude spectrum of acceleration will drop rapidly in the high frequency range. Anderson and Hough quantitatively studied the attenuation law of the acceleration Fourier spectrum above the corner frequency, and proposed the high-frequency attenuation parameter Kappa ( ) to describe this phenomenon. It has important applications in the field of engineering earthquakes, and is widely used by seismologists and engineers in the work of ground motion simulation and attenuation relations. Based on the Anderson method, this paper proposed a method of first smoothing the Fourier acceleration spectrum with a 0.4 Hz Parzen window and then fitting the solution value. The Fukushima earthquake data from Japan’s KiK-net seismic network on February 13, 2021 was used to verify the reliability of the method. The results showed that when the width of the Parzen window was set to 0.4 Hz, the calculation accuracy of the value was effectively improved.

Level set approach based on Parzen Window and floor of log for edge computing object segmentation in digital images

Scientific research into methodologies and algorithms to improve support for medical diagnoses remains in high demand on the agenda. Computer-aided diagnostic systems that use the Internet of Things (IoT) enable greater accessibility and integration between patients and specialists. One of the steps covered by IoT systems is the segmentation of Regions of Interest (ROI) in digital images. The challenge to segment these ROI in IoT systems is to apply methods that have low computational and storage costs with reliable results. Thus this work proposes the use of edge computing with a new approach based on active contours to target the ROI in medical images called FLog Parzen Level Set (FPLS). The method can be divided into two stages. First is the initialization of the seed point using Parzen Window and clusterization using Floor of Log. Second the growth and refinement of the region with probabilistic estimates of the regional contour using the Parzen Window. The proposed method was evaluated using the metrics of Accuracy, Precision, Sensitivity, Specificity, Matthews Coefficient Correlation, Hausdorff distance, Dice, and Jaccard Similarity Coefficient. The stable and satisfactory results can be highlighted despite the low computational times and costs. The proposed method was submitted to stroke, lung, and skin disease datasets. The proposed method achieved the fastest mean segmentation time of 1.64s for the three datasets used in this work. The method obtained the highest values for Sensitivity (98.57%), Accuracy (98.77%) and MCC (94.73%) in the stroke dataset, the lowest value for Hausdorff distance (4.24) in the lung dataset, and the highest Dice Coefficient value (92.49%) in the skin dataset. In conclusion, the proposed method is a promising tool for an edge computing system that segments regions of interest with the high precision of a level set and a fast convergence rate.

Sediment Classification of Acoustic Backscatter Image Based on Stacked Denoising Autoencoder and Modified Extreme Learning Machine

Acoustic backscatter data are widely applied to study the distribution characteristics of seabed sediments. However, the ghosting and mosaic errors in backscatter images lead to interference information being introduced into the feature extraction process, which is conducted with a convolutional neural network or auto encoder. In addition, the performance of the existing classifiers is limited by such incorrect information, meaning it is difficult to achieve fine classification in survey areas. Therefore, we propose a sediment classification method based on the acoustic backscatter image by combining a stacked denoising auto encoder (SDAE) and a modified extreme learning machine (MELM). The SDAE is used to extract the deep-seated sediment features, so that the training network can automatically learn to remove the residual errors from the original image. The MELM model, which integrates weighted estimation, a Parzen window and particle swarm optimization, is applied to weaken the interference of mislabeled samples on the training network and to optimize the random expression of input layer parameters. The experimental results show that the SDAE-MELM method greatly reduces mutual interference between sediment types, while the sediment boundaries are clear and continuous. The reliability and robustness of the proposed method are better than with other approaches, as assessed by the overall classification effect and comprehensive indexes.

Prediction of residual life of complex electromechanical systems based on small training samples

A model for predicting the residual life of complex electromechanical systems based on small training samples is presented. The developed algorithm involves the use of a probabilistic model of a complex electromechanical system. The statistical characteristics of the model parameters are determined by training samples, that is, by past experience of the functioning of this or similar system. The developed algorithm is based on recognition and forecasting algorithms based on a Wald’s sequential decision-making procedure, a modified V. S. Pugachev canonical decomposition, and an improved Parzen window density estimation algorithm.

Gaussian Mixture Model and Self-Organizing Map Neural-Network-Based Coverage for Target Search in Curve-Shape Area.

This article focuses on the target search problem in a curve-shape area using multiple unmanned aerial vehicles (UAVs), with the demand for obtaining the maximum cumulative detection reward, as well as the constraint of maneuverability and obstacle avoidance. First, the prior target probability map of the curve-shape area, generated by Parzen windows with Gaussian kernels, is approximated by the 1-D Gaussian mixture model (GMM) in order to extract some high-value curve segments corresponding to Gaussian components. Based on the parameterized curve segments from GMM, the self-organizing map (SOM) neural network is then established to achieve the coverage search. The step of winner neuron selection in SOM will prioritize and allocate the curve segments to UAVs, with the comprehensive consideration of multiple evaluation factors and allocation balance. The following step of neuron weight update will plan the UAV paths under the constraint of maneuverability and obstacle avoidance, using the modified Dubins guidance vector field. Finally, the good performance of GMM-SOM is evaluated on a coastline map.

Fast fully automatic skin lesions segmentation probabilistic with Parzen window

Cutaneous melanoma accounts for over 90% of all melanoma, causing up to 55,500 annual deaths. However, it is a potentially curable type of cancer. Since melanoma is potentially curable, the disease’s mortality rate is directly linked to late detection. This work proposes an approach that presents the balance between time and efficiency. This paper proposes the method of fast and automatic segmentation of skin lesions using probabilistic characteristics with the Parzen window (SPPW). The results obtained by the method were based on PH2 and ISIC datasets. The SPPW approach reached the following averages between the two datasets Specificity of 98.55%, Accuracy of 95.48%, Dice of 91.12%, Sensitivity of 88.45%, Mattheus of 87.86%, and Jaccard Index of 84.90%. The highlights of the proposed method are its short average segmentation time per image, and its metrics values, which are often higher than the ones obtained by other methods. Therefore, the SPPW method of segmentation is a quick, viable, and easily accessible option to aid in the diagnosis of diseased skin.

Feature subset and time segment selection for the classification of EEG data based motor imagery

The selection of feature subset and time segment is of great significance to the benefit of motor imagery classification. Hence, applying it in classification as well as itself alone has become an increasingly important research field in the brain-computer interface (BCI) systems. Most of the existing literatures only focus on binary-class classification situations in a fixed time segment. However, the modern BCI systems usually have to deal with more motor imagery classes. In this paper, we propose two Parzen window based methods to select the discriminative feature subset and subject-specific time segment. Further, we extend the proposed methods to multi-class issues. Finally, a soft Naive Bayesian classifier is designed to solve not only binary-class but also multi-class motor imagery problems. The proposed methods are validated on two well-known datasets, BCI competition IV dataset 2a and 2b. Experimental results reveal that the proposed methods achieve an improvement of 4.38 % for 2a and 2.54 % for 2b in comparison with the state-of-the-art methods, respectively. Meanwhile, both proposed methods achieve an improvement of 0.04 for 2a and 0.05 for 2b in kappa coefficient, respectively. Besides, the proposed multi-class methods would potentially contribute to the online BCI systems in practice.

Search for Static Target in Nonwide Area by AUV: A Prior Data-Driven Strategy

This article presents a prior data-driven strategy to solve the motion planning problem of autonomous underwater vehicle (AUV) for the target search mission. Compared with the widely studied cases over wide marine area, however, this article focuses on searching static target that exists in nonwide area. First, under the assumption that the nonwide area could be regarded as a combination of lines or curves, we utilize the Parzen windows theory with one-dimensional Gaussian kernels to generate the prior probability map of target based on the previous qualitative data. Then, we use the “global warming” method to extract some valuable subregions quantitatively, which will be visited by AUV in the optimal sequence. Finally, we verify the good performance of our method by an example of underwater pipeline inspection.

Machine learning distinguishes neurosurgical skill levels in a virtual reality tumor resection task.

This study outlines the first investigation of application of machine learning to distinguish "skilled" and "novice" psychomotor performance during a virtual reality (VR) brain tumor resection task. Tumor resection task participants included 23 neurosurgeons and senior neurosurgery residents as the "skilled" group and 92 junior neurosurgery residents and medical students as the "novice" group. The task involved removing a series of virtual brain tumors without causing injury to surrounding tissue. Originally, 150 features were extracted followed by statistical and forward feature selection. The selected features were provided to 4 classifiers, namely, K-Nearest Neighbors, Parzen Window, Support Vector Machine, and Fuzzy K-Nearest Neighbors. Sets of 5 to 30 selected features were provided to the classifiers. A working point of 15 premium features resulted in accuracy values as high as 90% using the Supprt Vector Machine. The obtained results highlight the potentials of machine learning, applied to VR simulation data, to help realign the traditional apprenticeship educational paradigm to a more objective model, based on proven performance standards. Graphical abstract Using several scenarios of virtual reality neurosurgical tumor resection together with machine learning classifiers to distinguish skill level.

Probabilistic Load Flow Based on Holomorphic Embedding, Kernel Density Estimator and Saddle Point Approximation Including Correlated Uncertainty Variables

Probabilistic Load Flow (PLF) methods have been extensively regarded due to the growth of uncertainty and renewable energies penetration in power systems. Given that, the current study aims to propose two novel probabilistic load flow methods based on holomorphic embedding method. Holomorphic method, unlike other iterative methods, introduces a nonlinear equation solver in an independent class, recursively. To that end, Kernel Density Estimator (KDE) and Saddle Point Approximation (SPA) methods are used to efficiently estimate probabilistic characteristics of load flow outputs. The correlations between random input variables with non-normal/normal probability distributions have been considered. The proposed algorithms have been examined on the modified IEEE 14- and 118-bus systems. Finally, to consider the uncorrelated and correlated conditions, the researcher compared the simulation results of the two proposed methods with some well-known and published probabilistic iterative load flow methods such as Monte Carlo Simulation (MCS), Parzen Window (PW), Diffusion method, and 2n+1 Point Estimate Method (PEM). MCS is always a dependable solution but time-consuming that make it useless for large power systems. In the present study, the accurate results obtained from MCS are regarded as a reference. The comparison of results shows high efficiency and low computational burden of proposed methods, and also the accuracy of these methods in density function estimation of output random variables.

Self-adaption vessel traffic behaviour recognition algorithm based on multi-attribute trajectory characteristics

Clustering analysis is commonly used for vessel traffic behaviour recognition. The clustering results reflect the characteristics of different vessel traffic behaviours, which can assist authorities in transportation management. However, there are two drawbacks to traditional clustering analysis. First, the similarity measures among different trajectories in a clustering analysis mainly focus on spatial differentiation and frequently do not consider motion features, such as speed and course. Second, the density-based spatial clustering of applications with noise (DBSCAN) algorithm, a famous clustering analysis method, is characterized by poor self-adaption, which means that it cannot autonomously determine the best parameters based on different sample sets. These issues limit the efficiency of clustering analysis. To address these problems, a new similarity measure using statistical methods and based on multi-attribute trajectory characteristics is proposed to reflect the similarity among different trajectories. The proposed measure considers the spatial and motion features of trajectories. Furthermore, the Parzen window (a non-parametric estimation method) and the multivariate Gaussian distribution are employed to generate self-adaptive strategies to determine the optimal parameters of the DBSCAN algorithm for a given sample set. Finally, numerical experiments are conducted to verify the effectiveness of the proposed similarity measure and improved DBSCAN algorithm. The clustering analysis results obtained with the proposed method could provide insights towards better monitoring and navigation advice to help improve marine managerial effectiveness and avoid maritime accidents.

Tracking Power System State Evolution with Maximum-correntropy-based Extended Kalman Filter

This paper develops a novel approach to track power system state evolution based on the maximum correntropy criterion, due to its robustness against non-Gaussian errors. It includes the temporal aspects on the estimation process within a maximum-correntropy-based extended Kalman filter (MCEKF), which is able to deal with both nonlinear supervisory control and data acquisition (SCADA) and phasor measurement unit (PMU) measurement models. By representing the behavior of the state variables with a nonparametric model within the kernel density estimation, it is possible to include abrupt state transitions as part of the process noise with non-Gaussian characteristics. Also, a novel strategy to update the size of Parzen windows in the kernel estimation is proposed to suppress the effects of suspect samples. By properly adjusting the kernel bandwidth, the proposed MCEKF keeps its accuracy during sudden load changes and contingencies, or in the presence of bad data. Simulations with IEEE test systems and the Brazilian interconnected system are carried out. The results show that the method deals with non-Gaussian noises in both the process and measurement, and provides accurate estimates of the system state under normal and abnormal conditions.

Feature selection for airborne LiDAR data filtering: a mutual information method with Parzon window optimization

ABSTRACT Filtering is one of the key steps for Digital Elevation Model (DEM) generation from airborne Light Detection and Ranging (LiDAR) data. Machine-learning-based filters have emerged as a class of filtering algorithms in recent years. Most existing studies mainly focus on feature generation due to limited available features a point cloud possesses. More than 30 features have been described in the existing literature. But most generated features are based on geometric information of points. Several redundant and irrelevant features may not necessarily improve the filtering accuracy. Hence, this paper proposes a feature-selection method using minimal-Redundancy-Maximal-Relevance (mRMR) combined with Parzen window optimization to deal with both discrete and continuous features. An optimal/suboptimal feature subset is constructed for machine-learning filters in various landscapes. Experimental results based on AdaBoost show that height-related features, particularly height itself, are of the greatest significance in both urban and rural scenes. Moreover, different subsets can be selected from the datasets of the two landscapes by our feature-selection strategy, which increases the data relevance for describing each geographical landscape. This study provides guidelines for the selection of optimal/suboptimal features for point cloud filtering based on machine-learning algorithms.

An IoT platform for the analysis of brain CT images based on Parzen analysis

Stroke or cerebrovascular accidents are among the three leading causes of death worldwide. Furthermore, strokes are a major cause of morbidity, hospitalizations, and acquired disabilities. Stroke diagnoses are usually made based on the set of symptoms exhibited and, more specifically, on the results of neuroimaging exams. Among the different types of neuroimaging, computed tomography (CT) is the most commonly used because it can show the extent and severity of the accident. Furthermore, CT exams are faster, more accessible than other systems, and are financially viable. Thus, due to its emergency nature, CAD systems that can analyze CT images are essential to obtain information that can accelerate the diagnosis and definition of appropriate treatment. This work presents a new feature extractor from CT images of the brain to assist in the detection and classification of strokes, called Parzen Analysis of Brain Tissue Densities. The proposed method can reduce the subjectivity present in the brain tissue bands using Parzen Window Estimation by calculating the probability that each pixel belongs to a preconfigured range. In addition, the method is fully integrated with an Internet of Things framework and can be used remotely to assist medical specialists in the diagnosis and treatment of strokes. The method presented promising results, reaching the highest values of accuracy, 98.41%, F1-Score, 97.61%, negative predictive value, 98.80%, and positive predictive value, 95.45%, among all the methods evaluated here. The results demonstrated the effectiveness of the method in extracting relevant features from brain CTs to describe if there is a stroke as well as determining its type.

Classification of EEG-based motor imagery BCI by using ECOC

AbstractAccuracy in identifying the subjects’ intentions for moving their different limbs from EEG signals is regarded as an important factor in the studies related to BCI. In fact, the complexity of motor-imagination and low amount of signal-to-noise ratio for EEG signal makes this identification as a difficult task. In order to overcome these complexities, many techniques such as various feature extraction methods, learning algorithms, and classifier schemes have been developed in this regard. However, conducting more research is necessary for improvement. The present study aimed to use an ensemble learning approach to improve the performance of MI-BCI systems. Therefore, filter bank common spatial pattern (FBCSP), as a well-known feature extraction method, was used to produce separable features from EEG signals. Accordingly, error correcting output codes (ECOC) was applied on several learning algorithms to classify four classes of motor imagery tasks. The proposed ECOC ensemble technique was tested on the data set 2a from BCI competition IV. Based on the results, the ECOC can lead to an improvement by using the naive Bayesian parzen window algorithm, compared to the winner algorithm of BCI competition IV, which is superior to other selected state of the art algorithms.

On evaluating the collaborative research areas: A case study

The growth of social networks is ever-increasing. Many available scientific publications evidence the interest of researchers in this area. Within a time span of eight years from 2011 to 2018, approximately 2600, 230, 150, and 110 scientific articles were published from the USA, Iran, Saudi Arabia, and Turkey, respectively around this area of research. To comprehensively survey all the sub-fields and interests within this research area, the present paper proposes a novel density-based method for finding topic descriptors from academic articles. By employing a robust to noise fuzzy clustering algorithm, the terms are clustered, and by utilizing a modified Parzen window, k topic descriptors from each cluster are extracted. Besides, an optimization problem has been designed to detect the similarity between word pairs. By conducting the experiments, the research priorities for four countries within this time span have been found. Moreover, the closeness of the research in developing countries to the developed country have been measured. The experimental results show that for four years, the research topics in Turkey were close to the research topics in the USA on average, and the research topics in Saudi Arabia were close to the USA topics during the past two years. Additionally, the experimental comparison of the proposed method with two clustering baselines indicates the superiority of the proposed method in terms of precision, recall, and accuracy.

ParaML: A Polyvalent Multicore Accelerator for Machine Learning

In recent years, machine learning (ML) techniques are proven to be powerful tools in various emerging applications. Traditionally, ML techniques are processed on general-purpose CPUs and GPUs, but their energy efficiencies are limited due to their excessive support for flexibility. As an efficient alternative to CPUs/GPUs, hardware accelerators are still limited as they often accommodate only a single ML technique (family). However, different problems may require different ML techniques, which implies that such accelerators may achieve poor learning accuracy or even be ineffective. In this paper, we present a polyvalent accelerator architecture integrated with multiple processing cores, called ParaML, which accommodates ten representative ML techniques, including $k$ -means, $k$ -nearest neighbors ( $k$ -NN), naive Bayes (NB), support vector machine (SVM), linear regression (LR), classification tree (CT), deep neural network (DNN), learning vector quantization (LVQ), parzen window (PW), and principal component analysis (PCA). Benefited from our thorough analysis on computational primitives and locality properties of different ML techniques, the single-core ParaML can perform up to 1056 GOP/s (e.g., additions and multiplications) in an area of 3.51 mm2 and consumes 596 mW only, estimated by ICC and PrimeTime PX with post-synthesis netlist, respectively. Compared with the NVIDIA K20M GPU (28-nm process), the single-core ParaML (65-nm process) is $1.21\times $ faster, and can reduce the energy by $137.93\times $ . We also compare the single-core ParaML with other accelerators. Compared with PRINS, single-core ParaML achieves $72.09\times $ and $2.57\times $ energy benefit for $k$ -NN and $k$ -means, respectively, and speeds up each query in $k$ -NN by $44.76\times $ . Compared with EIE, the single-core ParaML achieves $5.02\times $ speedup and $4.97\times $ energy benefit with $11.62\times $ less area when evaluating with dense DNN. Compared with TPU, the single-core ParaML achieves $2.45\times $ better power efficiency (5647 Gop/W versus 2300 Gop/W) with $321.36\times $ less area. Compared to the single-core version, the 8-core ParaML will further improve the speedup up to $3.98\times $ with an area of 13.44 mm2 and a power of 2036 mW.