Statistical Pattern Recognition System Research Articles

Technical diagnostic system in the maintenance of turbomachinery for ammonia synthesis in the process industries

Technical maintenance of machines and equipment in processing industry requires elaborate technical diagnostics systems to recognize the current state and forecast their future state. Creating such a system is a complex task due to multiple factors, with aging in aggressive exploitation environment being an important one. Statistical pattern recognition systems are very suitable to solve problems of technical diagnostics as they produce quantitative estimates of the states. We present the use of a hybrid Bayesian pattern recognition classifier that utilizes statistical and fuzzy paradigms and expresses the measurement information with four types of features (discrete, pseudo-discrete, multi-normal and independent continuous). It uses frequentist and subjective information (from training samples and expert opinion respectively) to identify the unknown parameters of the conditional likelihood density functions of each technical state. We discuss possible sources to collect learning information, and different methods to represent it. The classifier uses three different methods for parameter estimation of the conditional likelihood densities using data fusion. The classification is realised as a discriminant non-linear machine, which incorporates fuzzy approaches at different levels. We develop a novel algorithm for fault prediction without dynamic learning with four possible types of answers. A detailed example of technical diagnostics system for classification and prediction of states of turbomachinery for ammonia synthesis is presented. For the journal bearing diagnostics, we introduce modification of the hybrid Bayesian classifier using pseudo-priors to incorporate rule-based knowledge and improve the classification.

Development of fuzzy logic-based statistical process control chart pattern recognition system

This study developed a fuzzy logic system for online variation detection by recognizing patterns in statistical process control (SPC) charts. The system enables quality engineers to take corrective actions for an out-of-control manufacturing process. The fuzzy logic SPC system has three subsystems and is able to detect most common four patterns. Data go through subsystems in the absence of detective patterns. Otherwise, data will be eliminated when a particular pattern is recognized on the SPC chart. The user-friendly output interface reduces human analysis errors. The performance of the developed fuzzy logic system is tested by a simulated process.

Breath testing as a method for detecting lung cancer.

Early diagnosis of lung cancer is important due to high mortality in late stages of the disease. An ideal approach for population screening could be the breath analysis, due to its non-invasiveness, simplicity and cheapness. Using sensitive methods of analysis like gas chromatography/mass spectrometry in exhaled air of cancer patients were discovered some volatile organic compounds - possible candidates for cancer markers. However, these compounds were not specific for cancer cells. At the same time, integrative approaches used to analyze the exhaled breath have demonstrated high sensitivity and specificity of this method for lung cancer diagnosis. Such integrative approaches include detection of smell prints by electronic nose or integrated analysis of wide range of volatile organic compounds detected by gas chromatography/mass spectrometry or related methods. Modern statistical pattern recognition systems like logistic regression analysis, support vector machine or analysis by artificial neuronal network may improve diagnostic accuracy.

Handwritten Digit Recognition by Fourier-Packet Descriptors

Any statistical pattern recognition system includes a feature extraction component. For character patterns, several feature families have been tested, such as the Fourier-Wavelet Descriptors. We are proposing here a generalization of this family: the Fourier-Packet Descriptors. We have tested a set of 72 of these features on handwritten digits: the error rate was 2.44% with classifier 1NN for 19 features selected from the set and 1.72% with classifier SVM for all the set.

About the Authors

About the Authors

An improved scheme for online recognition of control chart patterns

This paper proposes two alternative schemes for the online recognition of control chart patterns (CCPs), namely: 1) a scheme based on direct continuous recognition; 2) a scheme based on 'recognition only when necessary'. The study focuses on recognition of six CCPs plotted on the Shewhart X-bar chart, namely, random, shift-up, shift down, trend-up, trend-down and cyclic. The artificial neural network (ANN) recogniser used was based on multilayer perceptrons (MLPs) architecture. The performance of the schemes was evaluated based on percentage correct recognition, average run lengths (ARL) and average recognition attempts (ARA). The findings suggest that the online recognition should be made only when necessary. Continuous recognition is not only wasteful, but also results in poorer results. The methodology proposed in this study is a step forward in realising a truly automated and intelligent online statistical process control chart pattern recognition system.

SVM-Based Detection of SAR Images in Partially Developed Speckle Noise

Support Vector Machine (SVM) is a statistical learning tool that was initially developed by Vapnik in 1979 and later developed to a more complex concept of structural risk minimization (SRM). SVM is playing an increasing role in applications to detection problems in various engineering problems, notably in statistical signal processing, pattern recognition, image analysis, and communication systems. In this paper, SVM was applied to the detection of SAR (synthetic aperture radar) images in the presence of partially developed speckle noise. The simulation was done for single look and multi-look speckle models to give a complete overlook and insight to the new proposed model of the SVM-based detector. The structure of the SVM was derived and applied to real SAR images and its performance in terms of the mean square error (MSE) metric was calculated. We showed that the SVM-detected SAR images have a very low MSE and are of good quality. The quality of the processed speckled images improved for the multi-look model. Furthermore, the contrast of the SVM detected images was higher than that of the original non-noisy images, indicating that the SVM approach increased the distance between the pixel reflectivity levels (the detection hypotheses) in the original images.

Real-Time Classification of Traffic Signs

A challenging real-time imaging problem is classifying video traffic signs in background clutter under rotation, scale, and translation invariant conditions. Normalized Gabor Wavelet Transform features from multi-resolution filters were originally biologically-based; however, optimized features proved more effective. Two whole image template matching techniques were unsuccessful. A statistical pattern recognition system recognized approximately 30% of the images for the original features and 50% for the optimized features; however, a multi-layer perceptron (mlp) detected over 70% of the images with the optimized features. The research demonstrated the possibility of a future automotive navigation aid which robustly collects sign images and classifies these images in real-time with a single Fast Fourier Transform (FFT), a bank of filters and a trained neural net.

Automated process monitoring using statistical pattern recognition techniques on X‐bar control charts

Quality control charts are statistical process control tools aimed at monitoring a (manufacturing) process to detect any deviations from normal operation and to aid in process diagnosis and correction. The information presented on the chart is a key to the successful implementation of a quality process correction system. Pattern recognition methodology has been pursued to identify unnatural behaviour on quality control charts. This approach provides the ability to utilize patterning information of the chart and to track back the root causes of process deviation, thus facilitating process diagnosis and maintenance. Presents analysis and development of a statistical pattern recognition system for the explicit identification of unnatural patterns on control charts. Develops a set of statistical pattern recognizers based on the likelihood ratio approach and on correlation analysis. Designs and implements a training algorithm to maximize the probability of identifying unnatural patterns, and presents a classification procedure for real‐time operation. Demonstrates the system performance using a set of newly defined measures, and obtained results based on extensive experiments illustrate the power and usefulness of the statistical approach for automating unnatural pattern detection on control charts.

A computer-based statistical pattern recognition for Doppler spectral waveforms of intracranial blood flow

A computer-based statistical pattern recognition system has been developed for the analysis of transcranial Doppler (TCD) spectral waveforms of the intracranial middle cerebral artery with varying degrees of increased intracranial pressure. This system extracts multidimensional features from TCD waveforms and performs a cluster analysis of those features. The system can automatically recognize the pattern of spectral waveform and classify it as a normal, abnormal, or borderline subclass of TCD spectral waveform. An optimum decision function was generated based on the Bayes Gaussian classifier. The accuracy of the Bayes Gaussian model the spectral waveforms reaches 100% by estimating posterior probability and using the resubstituting method of estimating misclassification in the training TCD data.

UNCERTAINTY REASONING IN ASTRONOMY--APPLICATIONS TO CLASSIFICATION AND OPTIMAL TELESCOPE SCHEDULING

ABSTRACT This thesis addresses two topics: the "selection of interesting objects from astronomical catalogues" and the "design of optimised observation schedules." Part I of the thesis deals with the question of how to optimally select interesting objects from astronomical databases. To this end we use a trainable statistical pattern recognition system. The chapter "Supervised and unsupervised classification --- The case of IRAS point sources" provides an short introduction to and review of supervised and un\\-supervised classification. The chapter "Mapping the extragalactic sky with IRAS PSC data" describes the selection of galaxy candidates from the IRAS Point Source Catalog. The distributions of the infrared colours are approximated as multivariate Gaussian distributions. The selection is carried out by statistical classifiers based on the maximum likelihood (ML) and maximum a posteriori (MAP) principles. The chapter "Young stellar objects in the IRAS Point Source Catalog" uses essentially the same methods for selecting young stellar object (YSO) candidates from the IRAS Point Source Catalog. In this case the information provided by the infrared colours alone is insufficient for a good discrimination of YSO candidates against other objects. However with the help of additional uncertain information provided by the spatial clustering properties of the objects on the sky, the selection can be significantly improved. The chapter "Statistical inferencing in the design of a sky-survey project" attempts to bridge the wide gap remaining between a selection of interesting candidates from some catalogue and the design of a suitably optimised strategy for follow-up observations. Part II of this thesis deals with the problem of how to create and optimise a long-term schedule for an astronomical telescope. Specifically the case of scheduling the Hubble Space Telescope (HST) observations is considered. It is shown how an artificial neural network with a novel stochastic dynamics can be used to efficiently solve the underlying constraint satisfaction problem (CSP). The chapter "The processing of HST observing programmes" introduces the HST long range planning problem and explains why the scheduling of pointed observations with this astronomical satellite in a low Earth orbit is difficult. The chapter "Connectionism and neural networks" provides some conceptual background on artificial neural networks. It is shown how a simple scheduling problem can be mapped onto the topology of an artificial neural network. Adding a suitable network dynamics turns such a network into an efficient problem solver. The chapter "A discrete stochastic neural network algorithm for constraint satisfaction problems" describes the novel "guarded discrete stochastic" (GDS) network algorithm, a relative of the standard Hopfield-Tank network, which is surprisingly efficient in finding solutions to large constraint satisfaction problems even on serial machines. The chapter "Scheduling with neural networks --- The case of Hubble Space Telescope" finally describes how the GDS network is integrated into Spike, the scheduling system developed at the Space Telescope Science Institute, Baltimore, for creating long range HST observing schedules. The methodology described here is quite general and is applicable to scheduling other ground-based or space-born astronomical observatories.

Computing with uncertainty

The role of and interaction between statistical, fuzzy, and neural-like models for certain problems associated with the three main areas of pattern recognition system design are discussed. Some questions concerning fuzzy sets are answered, and the design of fuzzy pattern recognition systems is reviewed. Pattern recognition, statistical pattern recognition and fuzzy pattern recognition systems are described. The use of computational neural-like networks in fuzzy pattern recognition is also discussed. >

Small sample size effects in statistical pattern recognition: recommendations for practitioners

The effects of sample size on feature selection and error estimation for several types of classifiers are discussed. The focus is on the two-class problem. Classifier design in the context of small design sample size is explored. The estimation of error rates under small test sample size is given. Sample size effects in feature selection are discussed. Recommendations for the choice of learning and test sample sizes are given. In addition to surveying prior work in this area, an emphasis is placed on giving practical advice to designers and users of statistical pattern recognition systems. >

Automatic solder joint inspection

The task of automating the visual inspection of pin-in-hole solder joints is addressed. Two approaches are explored: statistical pattern recognition and expert systems. An objective dimensionality-reduction method is used to enhance the performance of traditional statistical pattern recognition approaches by decorrelating feature data, generating feature weights, and reducing run-time computations. The expert system uses features in a manner more analogous to the visual clues that a human inspector would rely on for classification. Rules using these cues are developed, and a voting scheme is implemented to accumulate classification evidence incrementally. Both methods compared favorably with human inspector performance.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

A link between k-nearest neighbor rules and knowledge based systems by sequence analysis

We propose, in this paper, non-parametric decision rules used in a statistical pattern recognition system with incomplete knowledge. These decision rules are based on the examination of membership sequences of k-nearest neighbors. Beyond these parameters, we introduce the non-decision. To apply the decision rules, an expert system is used, which is also described in this paper. Lastly, a numerical and simulated example is given.

A Pattern Recognition Based Method for IC Failure Analysis

Random fluctuations which are inherent in the IC manufacturing process cause production yields to be significantly less than 100 percent. Yield drop is caused by two types of faults, catastrophic and parametric. This paper deals with the diagnosis of parametric faults which occur during the manufacturing of IC's and cause the yield to drop below some acceptable level. A statistical pattern recognition approach to IC failure analysis is developed in this paper. Identification of faults is based on the analysis of the joint probability density function of circuit performances. A number of alternative approaches to IC failure analysis are developed and their efficiency is discussed. Also a complete statistical pattern recognition system with learning capability is proposed and all the stages of its development are discussed. An efficient method for fault simulation which is based upon statistical process simulation is proposed. The performance of the algorithms proposed in this paper has been successfully verified for data collected from commercial fabrication processes and from computer simulation.

Classification of immature and mature cells of the neutrophil series using morphometrical parameters.

Quantitative morphological data of six classes of immature and mature cells of the neutrophil series of the bone marrow of normal persons were used for statistical classification experiments (myeloblasts, promyelocytes, myelocytes, metamyelocytes, bands and segments). On each cell, parameters were measured directly from the image or calculated from the shape of the density histogram or the counting densitogram using a Texture Analysis System (E. Leitz, Wetzlar, Germany). The parameters were analyzed with the interactive statistical pattern recognition system ISPAHAN. One half of the data were used as a learning set and the other half as the test set. The parameters were compared according to their performance in discrimination between the classes, alone and in combinations. Parameters not contributing to an improvement of the discrimination were disregarded. Eleven parameters were selected and used for classification by two different methods: a stepwise and a "one-shot" method. Stepwise classification resulted in a 79% correct classification rate. Most errors occurred between cell classes in neighboring stages of maturation. In 96% of all cases the computer classification was either in accordance with that of the technician or with a cell class of a neighboring maturation stage. One step classification by the computer was in agreement with the technicians in 82% of the cases. For 98% of the cells the computer classification was either in accordance with that of the technician or with a cell class of a neighboring maturation stage. The data set was collected by two technicians, operating independently. Differences in their interpretation of the maturation stage were found by comparing the performance of classifiers based on both cell samples. Since the images of the cells were not available for reexamination, the causes of disagreement in classification between the technicians and between computer and technicians could not be evaluated.

Machine Recognition of Abnormal Behavior in Nuclear Reactors

A multivariate statistical pattern recognition system for reactor noise analysis is presented. The basis of the system is a transformation for decoupling correlated variables and algorithms for inferring probability density functions. The system is adaptable to a variety of statistical properties of the data, and it has learning, tracking, updating, and dimensionality reduction capabilities. System design emphasizes control of the false-alarm rate. Its abilities to learn normal patterns and to recognize deviations from these patterns were evaluated by experiments at the Oak Ridge National Laboratory (ORNL) High-Flux Isotope Reactor. Power perturbations of less than 0.1 percent of the mean value in selected frequency ranges were readily detected by the pattern recognition system.

A Multivariate Statistical Pattern Recognition System for Reactor Noise Analysis

A multivariate statistical pattern recognition system for reactor noise analysis was developed. The basis of the system is a transformation for decoupling correlated variables and algorithms for inferring probability density functions. The system is adaptable to a variety of statistical properties of the data, and it has learning, tracking, and updating capabilities. System design emphasizes control of the false-alarm rate. The ability of the system to learn normal patterns of reactor behavior and to recognize deviations from these patterns was evaluated by experiments at the ORNL High-Flux Isotope Reactor (HFIR). Power perturbations of less than 0.1% of the mean value in selected frequency ranges were detected by the system.

Commission 6: Progress in Radio Waves and Transmission of Information: 2. Information Theory, Circuit Theory, and Computer‐Aided Design

As in previous summaries, the words ‘information theory’ are used in the wide sense to include such areas as (1) statistical communication theory, (2) detection theory, (3) pattern recognition and adaptive systems, (4) coding theory, and similar examples of the interaction of random processes and communication systems. For classification purposes, progress in information theory will be described under the four headings just listed.