Danton Diego Ferreira

Classification of lung sounds using higher-order statistics

BACKGROUND AND OBJECTIVE Lung sound auscultation is one of the most commonly used methods to evaluate respiratory diseases. However, the effectiveness of this method depends on the physicians training. If the physician does not have the proper training, he/she will be unable to distinguish between normal and abnormal sounds generated by the human body. Thus, the aim of this study was to implement a pattern recognition system to classify lung sounds. METHODS We used a dataset composed of five types of lung sounds: normal, coarse crackle, fine crackle, monophonic and polyphonic wheezes. We used higher-order statistics (HOS) to extract features (second-, third- and fourth-order cumulants), Genetic Algorithms (GA) and Fishers Discriminant Ratio (FDR) to reduce dimensionality, and k-Nearest Neighbors and Naive Bayes classifiers to recognize the lung sound events in a tree-based system. We used the cross-validation procedure to analyze the classifiers performance and the Tukeys Honestly Significant Difference criterion to compare the results. RESULTS Our results showed that the Genetic Algorithms outperformed the Fishers Discriminant Ratio for feature selection. Moreover, each lung class had a different signature pattern according to their cumulants showing that HOS is a promising feature extraction tool for lung sounds. Besides, the proposed divide-and-conquer approach can accurately classify different types of lung sounds. The classification accuracy obtained by the best tree-based classifier was 98.1% for classification accuracy on training, and 94.6% for validation data. CONCLUSIONS The proposed approach achieved good results even using only one feature extraction tool (higher-order statistics). Additionally, the implementation of the proposed classifier in an embedded system is feasible.

Improved Disturbance Detection Technique for Power-Quality Analysis

This paper presents an improved technique for detecting disturbances in voltage signals for power quality analysis. The main advantage of the proposed technique lies in its capability to detect disturbances when power frequency is time-varying. In addition, the technique is capable of detecting disturbances in frames corresponding to 1/64 of the fundamental component. Simulation results indicate that the proposed technique can offer improved performance in comparison with previous one.

ICA-Based Method for Power Quality Disturbance Analysis

This paper presents a new methodology based on Independent Component Analysis (ICA) for power quality disturbance analysis. The proposed methodology aims at analyzing power quality disturbances that appear as mixtures in the voltage signal. Such disturbances are commonly referred to as multiple disturbances. Results are obtained from both simulated and experimental data showing that disturbance classification higher than 97 % can be achieved. The results are attractive for practical applications in power quality systems.

Reducing electrocardiographic artifacts from electromyogram signals with independent component analysis

The aim of this work was to reduce ECG artifacts from surface electromyogram (EMG) signals collected from lumbar muscles with the blind source separation technique based on independent component analysis (ICA). Using four EMG signals collected above erector spinal lumbar muscles from 27 subjects, the proposed method fail in separating the sources. However, when considering a single channel of EMG and the same one time-shifted by one sample, the FastICA allowed reducing the signal to ECG noise ratio.

A direct approach for disturbance detection based on Principal Curves

This paper proposes a direct approach based on Principal Curves for power quality disturbance detection. The main advantages of the method lie in its low computational burden in the operational phase, its capability to detect disturbances when power frequency is time-varying and its good performance to signals exhibiting high noise levels. Simulation and experimental results show that the proposed method outperforms other methods in terms of detection rates and computational complexity.

Sub-harmonics detection and identification using higher order statistics

A large effort had been made for improving the quality of electric power in the last years. More often, the studies concern over methods and techniques to enhance monitoring systems with efficient fault detection and identification. This paper proposes a method based on higher order statistics for sub-harmonics detection and identification. Simulation results shown the accuracy and capability of the proposed technique, which furnishes a reasonable performance.

Real-time system for automatic classification of power quality disturbances

This paper presents an automatic system, implemented in LabVIEW, for real-time classification of electrical disturbances. Its basic structure can be listed in three stages: signal acquisition, feature extraction and final classification. The first one refers to the signal sampling by means of a monitoring embedded system and to filtering through a notch filter in order to divide the data into a fundamental component and an error signal. The RMS value and the second-order cumulants are extracted from the fundamental component and the error signal, respectively. The extracted features are then sent to the classification process, which is based on a decision tree constructed with perceptrons and a Bayesian classifier. It was possible to classify twenty classes of multiple and isolated disturbances. The results were satisfactory in which a classification accuracy of 98.47% was achieved for signals simulated through arbitrary waveform generator.

ICA-based method for power quality disturbance detection

This paper presents a method based on Independent Component Analysis (ICA) for power quality disturbance detection. The proposed method focuses on the detection of sag, swell, sinusoidal voltage fluctuation, fundamental frequency variations and harmonics in electric power signals. Suitable results were achieved from simulated signals in which the beginning and ending time of these disturbances were accurately detected. The results showed the proposed method can be used for disturbance segmentation as part of a power quality monitoring system.

Multidimensional monitoring for power quality disturbance detection

Power Quality (PQ) has emerged as an important research field in recent years. The development and increasing use of high power converters and the increase of nonlinear loads with high power cause unwanted changes in the electrical signal (current and voltage). These changes are called electrical disturbances. To understand such disturbances and investigate their causes, it is needed firstly detecting them. This work proposes a multidimensional approach, however with reduced computational complexity, for detecting PQ disturbances. The innovation of this work is the use of a deviation measure to quantify the PQ disturbances. The method performed well with high detection rates and low computational complexity.

Non-Intrusive Appliance Load Identification Based on Higher-Order Statistics

This paper presents a new method based on Higher-order Statistics for non-intrusive residential electrical load identification. Basically, the proposed method extracts cumulants of second and fourth order from the electric current signal of the residential electrical loads and presents these cumulants to a previously trained artificial neural network for classification. The neural network output identifies the residential electric load class of the processed signal. This study considered eleven different classes of residential electrical loads. Results were carried out from experimental electric signals and the achieved overall performance was over to 97%.