Statistical Pattern Recognition Techniques Research Articles

Multivariate Analysis of Single-Molecule Spectra: Surpassing Spectral Diffusion

The full exploitation of single-molecule spectroscopy in disordered systems is often hampered by spectral diffusion processes of the optical transitions due to structural fluctuations in the local environment of the probe molecule which leads to temporal averaging of the signal. Multivariate statistical pattern recognition techniques, originally developed for single-molecule cryoelectron microscopy, allow us to retrieve detailed information from optical single-molecule spectra. As an example, we present the phonon side band of the B800 excitations of the light-harvesting 2 (LH2) complex from Rhodospirillum molischianum, revealing the electron-phonon coupling strength for these transitions. The measured Debye-Waller factors, ranging from 0.4 to 0.9, fall in the regime of weak electron-phonon coupling.

Modified polytopic vector analysis to identify and quantify a dioxin dechlorination signature in sediments. 2. Application to the Passaic River.

Persistent contaminants such as dioxins have been documented to undergo dechlorination reactions in the laboratory; however, little is known about the importance of these reactions in the field. Polytopic vector analysis (PVA) is a statistical pattern recognition technique for multivariate data traditionally used to identify fingerprints of contaminant sources. A modified PVA algorithm with uncertainty analysis was used to model dechlorination fingerprints and sources. The technique was applied to 351 sediment core-derived dioxin samples from the lower reach of the Passaic River, New Jersey. A dechlorination fingerprint was identified with a highly positive 2,3,7,8-tetraCDD component and a highly negative heptaCDD component. The most important industrial source of 2,3,7,8-tetraCDD is a fingerprint related to 2,4,5-trichlorophenoxyacetic acid production. The dechlorination contribution to the data variance is 3.00 +/- 1.00%, corresponding to an average of 1.2 microg/kg of 2,3,7,8-tetraCDD per sample at the expense of heptaCDD. The possible occurrence of dechlorination was validated by comparing the local dechlorination contribution in the results to the value of the ratio 2,3,7,8-tetraCDD/total 2,3,7,8-PCDD, which indicates dechlorination in the laboratory. Bootstrap uncertainty analysis yielded the same dechlorination EM in 40% of the realizations. The results indicated that bootstrapping is an important statistical tool to quantify uncertainties with respect to the dechlorination EM and some of the source EMs.

Testing the performance of some nonparametric pattern recognition algorithms in realistic cases

The success obtained by Statistical Pattern Recognition in many disciplines is certainly related to the quality and availability of many data, normally distributed. However, in other disciplines, the data sets consist of few measurements, often binned, correlated, and not normally distributed. Usually, we do not even know which features have an influence on the process. The main goal of this paper is to evaluate the performance of some nonparametric Pattern Recognition algorithms when applied to such data. Finally we show the results of the application of the four nonparametric statistical pattern recognition techniques to real volcanological data.

Explicit pattern recognition models for speech perception

Optimal statistical classification of arbitrary input signals can be obtained, in principle, via a Bayesian classifier, given (perfect) knowledge of the distributions of signal properties for the set of target categories. At least for certain constrained problems, such as the perception of isolated vowels, simple (imperfect) statistical pattern recognition techniques can accurately predict human listeners’ performance. This paper sketches several relatively successful case studies of the application of static pattern recognition techniques to speech perception. (Static techniques require inputs of a fixed length, e.g., F1 and F2 for isolated vowels.) Real speech clearly requires dynamic pattern recognition, allowing inputs of arbitrary length. Certain such methods, such as dynamic programming and hidden Markov models, have been widely exploited in automatic speech recognition. The present paper will describe initial attempts to apply variants of such methods to the data from a perception experiment [T. Nearey and R. Smits, J. Acoust. Soc. Am. 111 (2002)] involving the perception of three (VCV) or four (VCCV) segment strings. Practical and conceptual problems in the application of such techniques to human perception will be discussed. [Work supported by SSHRC.]

THE DEVELOPMENT OF AUTOMATED PATTERN RECOGNITION AND STATISTICAL FEATURE ISOLATION TECHNIQUES FOR THE DIAGNOSIS OF RECIPROCATING MACHINERY FAULTS USING ACOUSTIC EMISSION

This work presents techniques which can be used for the diagnosis of faults in reciprocating machines. The techniques are generic in nature but the applications considered here concentrate on the interrogation of acoustic emission (AE) signals acquired from three types of reciprocating machine under normal running and with various fault conditions. Analysis of the signals in the time domain enables identification of machine timing, which can then be used to isolate automatically those parts of the signal associated with specific mechanical events. Various statistical feature isolation and pattern recognition techniques are then used on a selected time window of the signal to identify machine fault conditions. The applications considered demonstrate the generic nature of this approach, which could be applied to the monitoring of many types of machines and faults using a variety of sensor data. The ultimate aim of the work, of which this paper is a part, is to provide a means of on-line automatic monitoring of reciprocating machines using AE without recourse to any additional sensors.

Comparative analysis of different approaches to target differentiation and localization with sonar

This study compares the performances of different methods for the differentiation and localization of commonly encountered features in indoor environments. Differentiation of such features is of interest for intelligent systems in a variety of applications such as system control based on acoustic signal detection and identification, map building, navigation, obstacle avoidance, and target tracking. Different representations of amplitude and time-of-flight measurement patterns experimentally acquired from a real sonar system are processed. The approaches compared in this study include the target differentiation algorithm, Dempster–Shafer evidential reasoning, different kinds of voting schemes, statistical pattern recognition techniques ( k-nearest neighbor classifier, kernel estimator, parameterized density estimator, linear discriminant analysis, and fuzzy c-means clustering algorithm), and artificial neural networks. The neural networks are trained with different input signal representations obtained using pre-processing techniques such as discrete ordinary and fractional Fourier, Hartley and wavelet transforms, and Kohonen's self-organizing feature map. The use of neural networks trained with the back-propagation algorithm, usually with fractional Fourier transform or wavelet pre-processing results in near perfect differentiation, around 85% correct range estimation and around 95% correct azimuth estimation, which would be satisfactory in a wide range of applications.

Recognizing Correlated Patterns from Remotely Sensed Images: A New Learning Rule Based on Spin Glass Theory

Classification of remotely sensed images is a rich research field wherein techniques from conventional statistics to recent developments such as Artificial Neural Network, Fuzzy logic etc. has wide applications. Conventionally remotely sensed image classification referred to pixel classification based on broad categories such as vegetation and water bodies. With the availability of high-resolution imageries, shape analysis of macro structures contained in images becomes an important and difficult task. Although conventional statistical pattern recognition techniques give a reasonable result, Artificial neural network methods seem to be giving better results. In this paper, we give a survey of feed-forward neural network used for shape classification and a Hopfield model with an improved learning rule, for a typical shape analysis problem.

Ultrasonographic tissue characterization in monitoring tumor response to neoadjuvant chemotherapy in locally advanced breast cancer (work in progress).

A program of computer-assisted texture analysis was applied to evaluate its usefulness for objective description of changes in tumor architecture due to primary medical treatment in patients with locally advanced breast cancer. Changes in values of parameters of the statistical pattern recognition technique were compared to ultrasonographically depictable, subjectively recorded changes in echogenicity and echotexture (brightness, homogeneity) and reviewed with regard to histopathologic evaluation of tumor regression. Characteristic trends of defined quantitative texture parameters (mean gradient, mean gray value, contrast from the co-occurrence matrix) corresponded to visually depictable changes of the B-mode image and underlying histopathologic changes. The results indicate that quantitative texture analysis may aid in noninvasive monitoring of tumor response to neoadjuvant chemotherapy.

Robust methodology for the discrimination of brain tumors from in vivo magnetic resonance spectra (abstract only)

Magnetic resonance spectroscopy has become an important clinical tool, serving as a quantitative indicator of the chemical composition of tissue, whether sampled in vitro or in vivo. In particular, 1 H-MRS provides detailed biochemical information of tissues of interest and is increasingly used in tumors diagnosis and grading, where a number of automated decision support have shown promising results. Statistical pattern recognition (PR) techniques have an important role in the allocation of spectra to diagnostic classes because they utilise the multivariate nature of spectraHowever, the very large number of spectral components combined with significant noise effects, especially when acquired in vivo , pose substantial difficulties for their statistical analysis, and severely limit to generalise to future data the conclusion from studies using limited number of samples. Linear statistical models commonly be applied after the dimensionality of the spectrum has been reduced to a fraction of the number of samples available, and the reproducibility of the results obtained is strongly dependent on how this parsimony is achieved. Previous studies have passed on clinically identified variables of interest, typically area ratios or attempted to identify the most predictive frequency components. We propose a systematic variable selection methodology based on a well established statistical technique know as the bootstrap to select subset of variable for linear discriminant analysis (LDA) to classify different tissue types to class. This is complemented by the further use of the bootstrap to provide robust estimates of the accuracy of the discriminant models when they are applied to future data.The multivariate structure of MRS also makes it very difficult to visualise the separation between different tissues types. A 2-dimensional map of the class membership of the spectra also has considerable potential as a graphical aid for interactive decision support for differential diagnosis by clinicians. Therefore, we propose the application of a planar representation of the spectra, which preserves their relationship to the multivariate class means and variances. In this way, the discriminatory power for different selections of frequency components can be visually demonstrated. Finally, a reject option is introduced into the graphical display with the aim of minimising the risks of misclassification by the explicit visualisation of the overlap between spectra from tumor.

A two-stage process for automatic recognition of harbor seal vocalizations

Male harbor seals (Phoca vitulina) were continuously recorded for a year from an array of hydrophones in shallow water off the coast of central California. A two-stage automatic recognition system was used to extract sounds of interest. The first stage, fast but crude, processed the entire sound archive. It operated by: (1) making a spectrogram; (2) normalizing the spectrogram in several ways to remove some background noises and interfering sounds; and (3) detecting sounds in the 100–1000 Hz range with a minimum duration of 1 s. The second stage, slow but accurate, operated on the sounds extracted by the first stage and classified them as being either harbor seal roars or not. Classification was done by measuring a variety of acoustic characteristics—duration, frequency span, amplitude variation, etc.—in several frequency bands, and applying statistical pattern recognition techniques to the resulting feature vectors. Training data consisted of 1011 roar examples and 850 nonroar sounds. Recognition accuracy greater than 95% was achieved, with the principal errors occurring because of close resemblance between seal roars and breakingwaves. These results show that acoustic monitoring combined with automatic recognition can be a viable method for continuous monitoring of populations of wild animals. a)Currently at PMEL, 2115 S. E. OSU Dr., Newport, OR 97365.

Flat foot functional evaluation using pattern recognition of ground reaction data

Objective. Main purpose of this study was to apply quantitative gait analysis and statistical pattern recognition as clinical decision-making aids in flat foot diagnosis and post-surgery monitoring. Design. Statistical pattern recognition techniques were applied to discriminate between normal and flat foot populations through ground reaction force measurements; ground reaction forces time course was assumed as a sensible index of the foot function. Background. Gait analysis is becoming recognized as an important clinical tool in orthopaedics, in pre-surgery planning, post-surgery monitoring and in a posteriori evaluation of different treatment techniques. Statistical pattern recognition techniques have been utilized with success in this field to identify the most significant variables of selected motor functions in different pathologies, and to design classification rules and quantitative evaluation scores. Methods. Ground reaction forces were recorded during free speed barefoot walks on 28 healthy subjects, and 28 symptomatic flexible flat foot children selected for surgical intervention. A new feature selection algorithm, based on heuristic optimization, was applied to select the most discriminant ground reaction forces time samples. A two-stage pattern recognition system, composed by three linear feature extractors, one for each ground reaction force component, and a linear classifier, was designed to classify the feet of each subject using the selected features. The output of the classifier was used to define a functional score. Results. The classifier assigned the ground reaction force patterns performed by each subject into the right class with an estimated error of 15%, corresponding to an assignment error for each subject’s foot of 9%. The most discriminant ground reaction forces time samples selected are in full agreement with the pathophysiology of the symptomatic flexible flat foot. The obtained score was utilized to monitor the 1 and 2 years post-operative functional recovery of two differently treated subgroups of 32 flexible flat foot subjects. Conclusions. Statistical pattern recognition techniques are promising tools for clinical gait analysis; the obtained score provides important functional information that could be used as a further aid in the clinical evaluation of flat foot and different surgical treatment techniques. Relevance Symptomatic flexible flat foot surgical decision making is frequently difficult because of the lack of objective criteria to assess functional abnormalities of the foot/ankle complex. Gait analysis and statistical pattern recognition can give us parameters with which to characterize “functional” flat foot. Moreover, we can objectively follow up the recovery of the foot/ankle complex function after surgical treatment.

Use of parametric modelling and statistical pattern recognition in detection of awareness during general anaesthesia

Awareness is a rare but important complication of general anaesthesia. In its worst manifestation the patient is completely paralysed yet fully conscious and suffering the pain of the operative procedure. The sequelae from such an experience may be significant and lifelong. The paper describes a method, based on parametric modelling and statistical pattern recognition techniques, including neural networks, whereby awareness during general anaesthesia may be detected when present. Two systems are described, the first based solely on the use of the bispectrum, while the second makes use of both spectral and bispectral features. An evaluation on independent test sets shows that both systems have an average accuracy of > 80%, but the variation across individuals is less using the spectral–bispectral system (standard deviation of 16.4% compared with 20.5%). The spectral–bispectral system operates in near real time, requiring only 5 s of data to produce a new estimate of awareness. These estimates are obtained from the output of a trained neural network, which has as its input a set of features extracted from a single channel of electroencephalogram (EEG). The pre-processing of the data prior to input into the neural network is a critical component of the work, and it is here that parametric models have been extensively utilised. The spectral features are extracted from the EEG using a 1 s segment and a lattice filter as the primary model, while the bispectral features are extracted using a 5 s segment and a transversal filter as the underlying model.

Application of the rule-space procedure to language testing: examining attributes of a free response listening test

This study uses the rule-space methodology to explore the cognitive and linguistic attributes that underlie performance on an open-ended, short-answer, listening comprehension test. This is the first application of the methodology to language testing. The rule-space methodology is an adaptation of statistical pattern recognition techniques applied to the problem of diagnosing the cognitive attributes (knowledge, skills, abilities, strategies, etc.) underlying test performance. The methodology provides diagnostic information about the individual test-takers on each of these attributes. Based on a literature search, attribute candidates were identified which seemed likely to explain performance on the listening test. Two rule-space analyses were carried out, and the final attribute list had 15 attributes, with 14 interactions. With this, 96% of the test-takers were successfully classified into their respective latent knowledge states, and given scores on each attribute. This result suggests that the rule-space methodology can be used to identify attributes underlying performance on language tests. The study also provided useful information about the listening construct.

Automatic computerized radiographic identification of cephalometric landmarks

Computerized cephalometric analysis currently requires manual identification of landmark locations. This process is time-consuming and limited in accuracy. The purpose of this study was to develop and test a novel method for automatic computer identification of cephalometric landmarks. Spatial spectroscopy (SS) is a computerized method that identifies image structure on the basis of a convolution of the image with a set of filters followed by a decision method using statistical pattern recognition techniques. By this method, characteristic features are used to recognize anatomic structures. This study compared manual identification on a computer monitor and the SS automatic method for landmark identification on minimum resolution images (0.16 cm 2 per pixel). Minimum resolution (defined as the lowest resolution at which a cephalometric structure could be identified) was used to reduce computational time and memory requirements during this development stage of the SS method. Fifteen landmarks were selected on a set of 14 test images. The results showed no statistical difference ( p > 0.05) in mean landmark identification errors between manual identification on the computer display and automatic identification using SS. We conclude that SS shows potential for the automatic detection of landmarks, which is an important step in the development of a completely automatic cephalometric analysis. (Am J Orthod Dentofacial Orthop 1998;113:173–9)

Interpretation of Raman Spectra of Nitro-Containing Explosive Materials. Part II: The Implementation of Neural, Fuzzy, and Statistical Models for Unsupervised Pattern Recognition

The implementation of neural, fuzzy, and statistical models for the unsupervised pattern recognition and clustering of Fourier transform (FT)-Raman spectra of explosive materials is reported. In this work a statistical pattern recognition technique based on the concept of nearest-neighbors classification is described. Also the first application of both fuzzy clustering and a fuzzified Kohonen clustering network for the analysis of vibrational spectra is presented. Fuzzified Kohonen networks were found to perform as well as or better than the traditional fuzzy clustering technique. The unsupervised pattern recognition techniques, without the need for a priori structural information, yielded results which were comparable with those obtained by using a combination of a priori structural information and manual group-frequency analysis. This work demonstrates, via the use of a nitro-containing explosive data set, the utility of unsupervised pattern recognition techniques for the clustering, novelty detection, prototyping, and feature mapping of Raman spectra. The results of this work are directly applicable to the characterization of Raman spectra of explosives recorded with fiber-optic sampling.

Application of artificial neural networks to paleoceanographic data

Abstract Artificial neural networks are computer systems that have the ability to ‘learn’ a set of output, or target, vectors from a set of input vectors. Learning is achieved by self-adjustment of a set of parameters to minimize the error between a desired output and the actual network output. We have explored the potential of this approach in paleoceanography by application of a neural network algorithm to a problem involving prediction of sea surface-water temperatures from relative abundances of planktonic foraminifer species in the southern Indian Ocean. We employed a backpropagation (BP) network to assess how well it was able to predict the actual summer and winter surface-water temperatures. We compared the results with those obtained from statistical methods previously used for temperature predictions: Imbrie-Kipp Transfer Functions, the Modern Analog Technique, and Soft Independent Modelling of Class Analogy. The efficiency of predictions was tested using the Leaving One Out technique in which each of the observations in the data set is left out one at a time, while the remaining observations are used to generate a predictor. The accuracy of the predictor is then tested on the observation left out by comparison with its actual value. A set of tests using 1, 2, 3, 4, 5, and 10 neurons (processing elements) in a 3-layer BP network showed that a network with 3 neurons gave the smallest errors of prediction for both summer and winter temperatures, 0.71 and 0.76, respectively. Corresponding errors for the statistical pattern-recognition techniques ranged between 1.01 and 1.26 for summer temperatures and 1.05-1.13 for winter temperatures. Hence, predictions of paleotemperatures from new data on planktonic foraminifer relative abundances in the southern Indian Ocean may be made with a precision of ±0.7-0.8°C using the BP network and ±1.0–1.3°C using the statistical pattern-recognition procedures. The BP network was thus the most successful among the methods employed here for temperature predictions. Artificial neural networks may, therefore, be seen as a viable alternative to more conventional approaches to data analysis in paleoceanography.

Estimating the crowding level with a neuro-fuzzy classifier

This paper introduces a neuro-fuzzy system for the estimation of the crowding level in a scene. Monitoring the number of people present in a given indoor environment is a requirement in a variety of surveillance applications. In the present work, crowding has to be estimated from the image processing of visual scenes collected via a TV camera. A suitable preprocessing of the images, along with an ad hoc feature extraction process, is discussed. Estimation of the crowding level in the feature space is described in terms of a fuzzy decision rule, which relies on the membership of input patterns to a set of partially overlapping crowding classes, comprehensive of doubt classifications and outliers. A society of neural networks, either multilayer perceptrons or hyper radial basis functions, is trained to model individual class-membership functions. Integration of the neural nets within the fuzzy decision rule results in an overall neuro-fuzzy classifier. Important topics concerning the generalization ability, the robustness, the adaptivity and the performance evaluation of the system are explored. Experiments with real-world data were accomplished, comparing the present approach with statistical pattern recognition techniques, namely linear discriminant analysis and nearest neighbor. Experimental results validate the neuro-fuzzy approach to a large extent. The system is currently working successfully as a part of a monitoring system in the Dinegro underground station in Genoa, Italy.

Statistical pattern recognition technique to enhance anomalies in magnetic surveys

To classify and recognize objects or region of interest in an image, several methods in pattern recognition techniques were developed. Texture is one of the main characters used to classify and characterize an image whether it be an aerial photograph, a satellite image, a photomicrograph or, as in this study, a geophysical image. This paper describes the application of pattern recognition techniques to 2D geophysical data. The aim is to determine if they are useful to identify magnetic anomaly, especially with a very low signal/noise ratio. We utilized the method proposed by Haralick et al. (Haralick, R.M., Shanmugam, K. and Dinstein, I., 1973. Textural features for image classification. IEEE Trans. Syst. Man Cybern., 6-3: 610–621.) based on gray tone spatial dependence matrices; 14 features can be calculated to define the spatial relationship of pixels values; we verify that some of these features mark the boundaries of regions having a common statistical character and that match well with magnetic anomalies present in the image. Two synthetic models were utilized to test the discrimination capacity of the PR technique. Synthetic magnetic anomalies were computed by using the method proposed by Talwani (Talwani, M., 1965. Computation with the help of a digital computer of magnetic anomalies caused by bodies of arbitaray shape. Geophysics, 30: 797–817.). The synthetic images were altered with a random noise up to three times higher than anomaly amplitude. Then gray-tone spatial dependence matrix was applied for image classification. The same method has been finally used with experimental data collected during two magnetic surveys carried out in the same area for archaeological purpose. The results indicate that the method can be utilized to interpret geophysical data. Thus it can allow the anomalies detection and consequently facilitates geophysical data interpretation.

Classification by multiple-resolution statistical analysis with application to automated recognition of marine mammal sounds.

A multiple-resolution statistical pattern recognition technique for classification by supervised learning is developed and then applied to automated recognition of marine mammal sounds. The data to be classified may be either unprocessed or transformed, e.g., time series or time-frequency distributions of acoustic transients. Training data consist of samples previously grouped by a human expert into labeled sets; these sets are presumed to be associated with different "classes." The labeled sets are then characterized by occupancy statistics associated with a multiple-resolution, binary partition of the (unreduced) sample space. Classification of a new sample is performed by calculating a posteriori probabilities of membership of the new sample in each class, computed by Bayesian inference from the occupancy statistics of the associated labeled set. These a posteriori probabilities are calculated by a recursive algorithm that progresses from coarse to fine resolution in the sample space. The algorithm is implemented in a simple, highly efficient computer program. Automated classification of both time series and time-frequency distributions of marine-mammal vocalizations is demonstrated using a small number of labeled samples (approximately ten samples per class).

On neurobiological, neuro-fuzzy, machine learning, and statistical pattern recognition techniques

In this paper, we propose two new neuro-fuzzy schemes, one for classification and one for clustering problems. The classification scheme is based on Simpson's fuzzy min-max method (1992, 1993) and relaxes some assumptions he makes. This enables our scheme to handle mutually nonexclusive classes. The neuro-fuzzy clustering scheme is a multiresolution algorithm that is modeled after the mechanics of human pattern recognition. We also present data from an exhaustive comparison of these techniques with neural, statistical, machine learning, and other traditional approaches to pattern recognition applications. The data sets used for comparisons include those from the machine learning repository at the University of California, Irvine. We find that our proposed schemes compare quite well with the existing techniques, and in addition offer the advantages of one-pass learning and online adaptation.