Michael E. Zervakis

Classification of washing machines vibration signals using discrete wavelet analysis for feature extraction

Wavelets provide a powerful tool for nonstationary signal analysis. In vibration monitoring, the occurrence of occasional transient disturbances makes the recorded signal nonstationary, especially during the start-up of an engine. Through the wavelet analysis, transients can be decomposed into a series of wavelet components, each of which is a time-domain signal that covers a specific octave frequency band. Disturbances of small extent (duration) are amplified relative to the rest of the signal when projected to similar size wavelet bases and, thus, they can be easily detected in the corresponding frequency band. This paper presents a new method for extracting features in the wavelet domain and uses them for classification of washing machines vibration transient signals. The discrete wavelet transform (DWT), in conjunction with statistical digital signal processing techniques, is used for feature extraction. The Karhunen Loeve transform (KLT) is used for feature reduction and decorrelation of the feature vectors. The Euclidean, Mahalanobis, and Bayesian distance classifiers, the learning vector quantization (LVQ) classifier, and the fuzzy gradient classifier are used for classification of the resulting feature space. Classification results are illustrated and compared for the rising part of vibration velocity signals of a variety of real washing machines with various defects.

High-order neural network structure selection for function approximation applications using genetic algorithms

Neural network literature for function approximation is by now sufficiently rich. In its complete form, the problem entails both parametric (i.e., weights determination) and structural learning (i.e., structure selection). The majority of works deal with parametric uncertainty assuming knowledge of the appropriate neural structure. In this paper we present an algorithmic approach to determine the structure of high order neural networks (HONNs), to solve function approximation problems. The method is based on a genetic algorithm (GA) and is equipped with a stable update law to guarantee parametric learning. Simulation results on an illustrative example highlight the performance and give some insight of the proposed approach.

Artificial neural networks for discriminating pathologic from normal peripheral vascular tissue

The identification of the state of human peripheral vascular tissue by using artificial neural networks is discussed in this paper. Two different laser emission lines (He-Cd, Ar+) are used to excite the chromophores of tissue samples. The fluorescence spectrum obtained, is passed through a nonlinear filter based on a high-order (HO) neural network neural network (NN) [HONN] whose weights are updated by stable learning laws, to perform feature extraction. The values of the feature vector reveal information regarding the tissue state. Then a classical multilayer perceptron is employed to serve as a classifier of the feature vector, giving 100% successful results for the specific data set considered. Our method achieves not only the discrimination between normal and pathologic human tissue, but also the successful discrimination between the different types of pathologic tissue (fibrous, calcified). Furthermore, the small time needed to acquire and analyze the fluorescence spectra together with the high rates of success, proves our method very attractive for real-time applications.

A class of robust entropic functionals for image restoration

This paper considers the concept of robust estimation in regularized image restoration. Robust functionals are employed for the representation of both the noise and the signal statistics. Such functionals allow the efficient suppression of a wide variety of noise processes and permit the reconstruction of sharper edges than their quadratic counterparts. A new class of robust entropic functionals is introduced, which operates only on the high-frequency content of the signal and reflects sharp deviations in the signal distribution. This class of functionals can also incorporate prior structural information regarding the original image, in a way similar to the maximum information principle. The convergence properties of robust iterative algorithms are studied for continuously and noncontinuously differentiable functionals. The definition of the robust approach is completed by introducing a method for the optimal selection of the regularization parameter. This method utilizes the structure of robust estimators that lack analytic specification. The properties of robust algorithms are demonstrated through restoration examples in different noise environments.

A hybrid neural network/genetic algorithm approach to optimizing feature extraction for signal classification

In this paper, a hybrid neural network/genetic algorithm technique is presented, aiming at designing a feature extractor that leads to highly separable classes in the feature space. The application upon which the system is built, is the identification of the state of human peripheral vascular tissue (i.e., normal, fibrous and calcified). The system is further tested on the classification of spectra measured from the cell nuclei in blood samples in order to distinguish normal cells from those affected by Acute Lymphoblastic Leukemia. As advantages of the proposed technique we may encounter the algorithmic nature of the design procedure, the optimized classification results and the fact that the system performance is less dependent on the classifier type to be used.

A Bayesian framework for multilead SMD post-placement quality inspection

In this paper, a novel framework is proposed to inspect the placement quality of surface mount technology devices (SMDs), immediately after they have been placed in wet solder paste on a printed circuit board (PCB). The developed approach involves the indirect measurement of each lead displacement with respect to its ideal position, centralized on its pad region. This displacement is inferred from area measurements on the raw image data of the lead region through a classification process. To increase the accuracy in the computation of the lead displacement, we introduce a combined classification/estimation process, in which the individual lead displacement classifications are viewed as measurements (or observations) of the same physical quantity i.e., the displacement of the entire component as a rigid body. Certain geometric relations connecting lead shifts to component displacement are also derived. Employing these relations we can infer a new refined measurement of the shift of each individual lead, a quantity crucial to the calculation of the quality measures. Experimental results highlight the potential of the developed algorithm.

Potential dementia biomarkers based on the time-varying microstructure of sleep EEG spindles

The time-varying microstructure of sleep EEG spindles may have clinical significance in dementia studies. In this work, the sleep spindle is modeled as an AM-FM signal and parameterized in terms of six parameters, three quantifying the instantaneous envelope (IE) and three quantifying the instantaneous frequency (IF) of the spindle model. The IE and IF waveforms of sleep spindles from patients with dementia and normal controls were estimated using the time-frequency technique of complex demodulation (CD). Sinusoidal curve-fitting using a matching pursuit (MP) approach was applied to the IE and IF waveforms for the estimation of the six model parameters. Specific differences were found in sleep spindle instantaneous frequency dynamics between spindles from dementia subjects and spindles from controls.

A wavelet-domain algorithm for denoising in the presence of noise outliers

A wavelet domain robust denoising algorithm is presented, which efficiently removes both Gaussian as well as Gaussian mixed with impulse noise. Several wavelet domain operators are developed which help in the denoising process. The superiority of the new algorithm is firmly established by simulation over a variety of images.

Strengths and Weaknesses of 1.5T and 3T MRS Data in Brain Glioma Classification

Although magnetic resonance spectroscopy (MRS) methods of 1.5Tesla (T) and 3T have been widely applied during the last decade for noninvasive diagnostic purposes, only a few studies have been reported on the value of the information extracted in brain cancer discrimination. The purpose of this study is threefold. First, to show that the diagnostic value of the information extracted from two different MRS scanners of 1.5T and 3T is significantly influenced in terms of brain gliomas discrimination. Second, to statistically evaluate the discriminative potential of publicly known metabolic ratio markers, obtained from these two types of scanners in classifying low-, intermediate-, and high-grade gliomas. Finally, to examine the diagnostic value of new metabolic ratios in the discrimination of complex glioma cases where the diagnosis is both challenging and critical. Our analysis has shown that although the information extracted from 3T MRS scanner is expected to provide better brain gliomas discrimination; some factors like the features selected, the pulse-sequence parameters, and the spectroscopic data acquisition methods can influence the discrimination efficiency. Finally, it is shown that apart from the bibliographical known, new metabolic ratio features such as N-acetyl aspartate/S, Choline/S, Creatine/S , and myo-Inositol/S play significant role in gliomas grade discrimination.

Statistical analysis on polarimetric study of lung cancer cells

The objective of this study is the discrimination and characterization of different lung cancer monoline cells using statistical analysis of polarimetric backscattered signals. The main aspect of this study is the use of the Welchs t-test and the p-value statistics as a representative metric for discriminating distributions based on their mean and standard deviation. The outcome of this study indicates that enhanced discrimination of lung cancer samples can be obtained based on their t-test values between different cancer samples for different geometries.