Noisy texture classification: A higher-order statistics approach

In this paper, a novel texture classification scheme using higher-order statistics (HOS) functions as discriminating features is proposed. It is well known that such statistical parameters are insensitive to additive Gaussian noise. In particular, third-order statistical parameters, i.e. third-order cumulants and bispectrum, are insensitive to any symmetrically distributed noise, and also exhibit the capability of better characterizing non-Gaussian signals. By exploiting these HOS properties, it is possible to devise a robust method for classifying textures affected by noise with different distributions and even with very low signal-to-noise ratios.     

特征与分类算法在基于肌音信号的头部运动分类中的对比研究

提取颈部肌肉的肌音（Mechanomyography, MMG）信号时域、时-频域和非线性动力学的15个常见特征,按照其性质分为5个特征集,并选择其中一部分构建高维特征矢量后进行主成分分析（Principal component analysis, PCA）降维处理,应用于头部动作的模式识别研究中。分别采用支持向量机(Support vector machine, SVM)、K近邻(K-nearest neighbor,KNN)和线性判别分析(Linear discriminant analysis, LDA)3种分类器,对6种头部动作（低头、抬头、左摆头、右摆头、左转头和右转头）的MMG信号进行分类。实验结果表明,选用时域、时-频域和非线性动力学特征组合的方式,以及使用SVM作为分类器,可使各类动作的分类精度均达到80%以上,从而获得相对较高的准确率。     

Differentiating types of muscle movements using a wavelet based fuzzy clustering neural network

Abstract: The electromyographic signals observed at the surface of the skin are the sum of many small action potentials generated in the muscle fibres. After the signals are processed, they can be used as a control source of multifunction prostheses. The myoelectric signals are represented by wavelet transform model parameters. For this purpose, four different arm movements (elbow extension, elbow flexion, wrist supination and wrist pronation) are considered in studying muscle contraction. Wavelet parameters of myoelectric signals received from the muscles for these different movements were used as features to classify the electromyographic signals in a fuzzy clustering neural network classifier model. After 1000 iterations, the average recognition percentage of the test was found to be 97.67% with clustering into 10 features. The fuzzy clustering neural network programming language was developed using Pascal under Delphi.     

Combining discriminative spatiotemporal features for daily life activity recognition using wearable motion sensing suit

Motion sensing plays an important role in the study of human movements, motivated by a wide range of applications in different fields, such as sports, health care, daily activity, action recognition for surveillance, assisted living and the entertainment industry. In this paper, we describe how to classify a set of human movements comprising daily activities using a wearable motion capture suit, denoted as FatoXtract. A probabilistic integration of different classifiers recently proposed is employed herein, considering several spatiotemporal features, in order to classify daily activities. The classification model relies on the computed confidence belief from base classifiers, combining multiple likelihoods from three different classifiers, namely Naïve Bayes, artificial neural networks and support vector machines, into a single form, by assigning weights from an uncertainty measure to counterbalance the posterior probability. In order to attain an improved performance on the overall classification accuracy, multiple features in time domain (e.g., velocity) and frequency domain (e.g., fast Fourier transform), combined with geometrical features (joint rotations), were considered. A dataset from five daily activities performed by six participants was acquired using FatoXtract. The dataset provided in this work was designed to be extremely challenging since there are high intra-class variations, the duration of the action clips varies dramatically, and some of the actions are quite similar (e.g., brushing teeth and waving, or walking and step). Reported results, in terms of both precision and recall, remained around 85 %, showing that the proposed framework is able to successfully classify different human activities.     

Identification of EMG signals using discriminant analysis and SVM classifier

The electromyography (EMG) signal is a bioelectrical signal variation, generated in muscles during voluntary or involuntary muscle activities. The muscle activities such as contraction or relaxation are always controlled by the nervous system. The EMG signal is a complicated biomedical signal due to anatomical/physiological properties of the muscles and its noisy environment. In this paper, a classification technique is proposed to classify signals required for a prosperous arm prosthesis control by using surface EMG signals. This work uses recorded EMG signals generated by biceps and triceps muscles for four different movements. Each signal has one single pattern and it is essential to separate and classify these patterns properly. Discriminant analysis and support vector machine (SVM) classifier have been used to classify four different arm movement signals. Prior to classification, proper feature vectors are derived from the signal. The feature vectors are generated by using mean absolute value (MAV). These feature vectors are provided as inputs to the identification/classification system. Discriminant analysis using five different approaches, classification accuracy rates achieved from very good (98%) to poor (96%) by using 10-fold cross validation. SVM classifier gives a very good average accuracy rate (99%) for four movements with the classification error rate 1%. Correct classification rates of the applied techniques are very high which can be used to classify EMG signals for prosperous arm prosthesis control studies.     

EMG feature evaluation for improving myoelectric pattern recognition robustness

In pattern recognition-based myoelectric control, high accuracy for multiple discriminated motions is presented in most of related literature. However, there is a gap between the classification accuracy and the usability of practical applications of myoelectric control, especially the effect of long-term usage. This paper proposes and investigates the behavior of fifty time-domain and frequency-domain features to classify ten upper limb motions using electromyographic data recorded during 21 days. The most stable single feature and multiple feature sets are presented with the optimum configuration of myoelectric control, i.e. data segmentation and classifier. The result shows that sample entropy (SampEn) outperforms other features when compared using linear discriminant analysis (LDA), a robust classifier. The averaged test classification accuracy is 93.37%, when trained in only initial first day. It brings only 2.45% decrease compared with retraining schemes. Increasing number of features to four, which consists of SampEn, the fourth order cepstrum coefficients, root mean square and waveform length, increase the classification accuracy to 98.87%. The proposed techniques achieve to maintain the high accuracy without the retraining scheme. Additionally, this continuous classification allows the real-time operation.     

Linear and nonlinear analysis of normal and CAD-affected heart rate signals

Coronary artery disease (CAD) is one of the dangerous cardiac disease, often may lead to sudden cardiac death. It is difficult to diagnose CAD by manual inspection of electrocardiogram (ECG) signals. To automate this detection task, in this study, we extracted the heart rate (HR) from the ECG signals and used them as base signal for further analysis. We then analyzed the HR signals of both normal and CAD subjects using (i) time domain, (ii) frequency domain and (iii) nonlinear techniques. The following are the nonlinear methods that were used in this work: Poincare plots, Recurrence Quantification Analysis (RQA) parameters, Shannon entropy, Approximate Entropy (ApEn), Sample Entropy (SampEn), Higher Order Spectra (HOS) methods, Detrended Fluctuation Analysis (DFA), Empirical Mode Decomposition (EMD), Cumulants, and Correlation Dimension. As a result of the analysis, we present unique recurrence, Poincare and HOS plots for normal and CAD subjects. We have also observed significant variations in the range of these features with respect to normal and CAD classes, and have presented the same in this paper. We found that the RQA parameters were higher for CAD subjects indicating more rhythm. Since the activity of CAD subjects is less, similar signal patterns repeat more frequently compared to the normal subjects. The entropy based parameters, ApEn and SampEn, are lower for CAD subjects indicating lower entropy (less activity due to impairment) for CAD. Almost all HOS parameters showed higher values for the CAD group, indicating the presence of higher frequency content in the CAD signals. Thus, our study provides a deep insight into how such nonlinear features could be exploited to effectively and reliably detect the presence of CAD.     

Complex networks approach for EEG signal sleep stages classification

Sleep stage scoring is a challenging task. Most of existing sleep stage classification approaches rely on analysing electroencephalography (EEG) signals in time or frequency domain. A novel technique for EEG sleep stages classification is proposed in this paper. The statistical features and the similarities of complex networks are used to classify single channel EEG signals into six sleep stages. Firstly, each EEG segment of 30 s is divided into 75 sub-segments, and then different statistical features are extracted from each sub-segment. In this paper, feature extraction is important to reduce dimensionality of EEG data and the processing time in classification stage. Secondly, each vector of the extracted features, which represents one EEG segment, is transferred into a complex network. Thirdly, the similarity properties of the complex networks are extracted and classified into one of the six sleep stages using a k-means classifier. For further investigation, in the statistical features extraction phase two statistical features sets are tested and ranked based on the performance of the complex networks. To investigate the classification ability of complex networks combined with k-means, the extracted statistical features were also forwarded to a k-means and a support vector machine (SVM) for comparison. We also compare the proposed method with other existing methods in the literature. The experimental results show that the proposed method attains better classification results and a reasonable execution time compared with the SVM, k-means and the other existing methods. The research results in this paper indicate that the proposed method can assist neurologists and sleep specialists in diagnosing and monitoring sleep disorders.     

Classifying watermelon ripeness by analysing acoustic signals using mobile devices

This work addresses the problem of distinguishing between ripe and unripe watermelons using mobile devices. Through analysing ripeness-related features extracted by thumping watermelons, collecting acoustic signals by microphones on mobile devices, our method can automatically identify the ripeness of watermelons. This is possible in real time, making use of machine learning techniques to provide good accuracy. We firstly collect a training dataset comprising acoustic signals generated by thumping both ripe and unripe watermelons. Audio signal analysis on this helps identify features related to watermelon ripeness. These features are then used to construct a classification model for future signals. Based on this, we developed a crowdsourcing application for Android which allows users to identify watermelon ripeness in real time while submitting their results to us allowing continuous improvement of the classification model. Experimental results show that our method is currently able to correctly classify ripe and unripe watermelons with an overall accuracy exceeding 89 %.     

A novel Parkinson’s Disease Diagnosis Index using higher-order spectra features in EEG signals

Higher-order spectra (HOS) is an efficient feature extraction method used in various biomedical applications such as stages of sleep, epilepsy detection, cardiac abnormalities, and affective computing. The motive of this work was to explore the application of HOS for an automated diagnosis of Parkinson’s disease (PD) using electroencephalography (EEG) signals. Resting-state EEG signals collected from 20 PD patients with medication and 20 age-matched normal subjects were used in this study. HOS bispectrum features were extracted from the EEG signals. The obtained features were ranked using t value, and highly ranked features were used in order to develop the PD Diagnosis Index (PDDI). The PDDI is a single value, which can discriminate the two classes. Also, the ranked features were fed one by one to the various classifiers, namely decision tree (DT), fuzzy K-nearest neighbor (FKNN), K-nearest neighbor (KNN), naive bayes (NB), probabilistic neural network (PNN), and support vector machine (SVM), to choose the best classifier using minimum number of features. We have obtained an optimum mean classification accuracy of 99.62%, mean sensitivity and specificity of 100.00 and 99.25%, respectively, using the SVM classifier. The proposed PDDI can aid the clinicians in their diagnosis and help to test the efficacy of drugs.

     

Clustering MIT–BIH arrhythmias with Ant Colony Optimization using time domain and PCA compressed wavelet coefficients

In this paper, Ant Colony Optimization (ACO) based clustering analysis of ECG arrhythmias taken from the MIT–BIH Arrhythmia Database is proposed. Both time domain and discrete wavelet transform (DWT) based frequency domain features are used in the analysis. Since the number of wavelet coefficients are huge amount as compared to the time domain parameters, Principal Component Analysis (PCA) based compression is applied on them in order to decrease their number to the number of time domain features. Then, the reduced numbers of frequency parameters are combined with the time domain features, in order to get the total feature sets. Different types of feature sets are tried and the classification results are compared. These are: time domain feature set, frequency domain feature set and the mixture of them. A neural network algorithm is developed in parallel to verify and measure the ACO classifier's success. Moreover, linear discriminant analysis (LDA) is used to show the effect of clustering on the system's results. The method is tested with MIT–BIH database to classify normal beats and five different critical and having vital importance arrhythmia types. Chosen six classes are normal sinus rhythm, premature ventricular contraction (PVC), atrial premature contraction (APC), right bundle branch block (RBBB), ventricular fusion (F) and fusion (f). Comparison results indicate that the mixture feature set gave a better success for the classification.     

表面肌电信息融合与动作分类

提出基于多个特征域信息融合的方法,进一步去除不确定性、提高表面肌电分类准确率.选择的表面肌电特征参数分别为时域绝对值积分、AR模型系数和线性倒谱系数.待辨识的6类手部动作肌电信号经各特征域变换,提取特征矢量后由BP神经网络分类,根据D-S证据理论对各分类器分类结果进行证据累积,并得到最终分类结果.实验结果表明,动作分类准确率高于传统的单特征集单分类器的分类方法,且训练、分类效率高于结构化神经网络特征融合方法.     

Performance Analysis of Machine Learning Algorithms for Classifying Hand Motion-Based EEG Brain Signals

Brain-computer interfaces (BCIs) records brain activity using electroencephalogram (EEG) headsets in the form of EEG signals; these signals can be recorded, processed and classified into different hand movements, which can be used to control other IoT devices. Classification of hand movements will be one step closer to applying these algorithms in real-life situations using EEG headsets. This paper uses different feature extraction techniques and sophisticated machine learning algorithms to classify hand movements from EEG brain signals to control prosthetic hands for amputated persons. To achieve good classification accuracy, denoising and feature extraction of EEG signals is a significant step. We saw a considerable increase in all the machine learning models when the moving average filter was applied to the raw EEG data. Feature extraction techniques like a fast fourier transform (FFT) and continuous wave transform (CWT) were used in this study; three types of features were extracted, i.e., FFT Features, CWT Coefficients and CWT scalogram images. We trained and compared different machine learning (ML) models like logistic regression, random forest, k-nearest neighbors (KNN), light gradient boosting machine (GBM) and XG boost on FFT and CWT features and deep learning (DL) models like VGG-16, DenseNet201 and ResNet50 trained on CWT scalogram images. XG Boost with FFT features gave the maximum accuracy of 88%.     

Rotating machine fault detection based on HOS and artificial neural networks

In order to identify the faults of rotating machinery, classification process can be divided into two stages: one is the signal preprocessing and the feature extraction; the other is the recognition process. In the preprocessing and feature extraction stage, the higher-order statistics (HOS) is used to extract features from the vibration signals. In the recognition process, two kinds of neural network classifier are used to evaluate the classification results. These two classifiers are self-organizing feature mapping (SOM) network for collecting data at the initial stage and learning vector quantization (LVQ) network at the identification stage. The experimental results obtained from HOS as preprocessor to extract the features of fault are clearer than those obtained from the power spectrum. In addition, the recognizable rate by using either SOM or LVQ as classifiers is 100%.     

Toward improved control of prosthetic fingers using surface electromyogram (EMG) signals

A fundamental component of many modern prostheses is the myoelectric control system, which uses the electromyogram (EMG) signals from an individual’s muscles to control the prosthesis movements. Despite the extensive research focus on the myoelectric control of arm and gross hand movements, more dexterous individual and combined fingers control has not received the same attention. The main contribution of this paper is an investigation into accurately discriminating between individual and combined fingers movements using surface EMG signals, so that different finger postures of a prosthetic hand can be controlled in response. For this purpose, two EMG electrodes located on the human forearm are utilized to collect the EMG data from eight participants. Various feature sets are extracted and projected in a manner that ensures maximum separation between the finger movements and then fed to two different classifiers. The second contribution is the use of a Bayesian data fusion postprocessing approach to maximize the probability of correct classification of the EMG data belonging to different movements. Practical results and statistical significance tests prove the feasibility of the proposed approach with an average classification accuracy of ≈90% across different subjects proving the significance of the proposed fusion scheme in finger movement classification.     

基于经验模态分解和SVM的脑电信号分类方法

脑电信号的非线性、非平稳性造成对运动想象脑电信号的分类识别存在特征提取困难、可区分性低以及分类识别性能差等问题。本文提出一种基于经验模态分解(Empirical Mode Decomposition, EMD)和支撑向量机(Support Vector Machine, SVM)的运动想象脑电信号分类方法,充分利用EMD算法在处理非线性、非平稳信号的自适应性以及SVM在小样本条件的高识别性能和强泛化能力。首先利用EMD算法将C3、C4导联信号分解为一系列本征模函数(Intrinsic Mode Function, IMF),然后从IMF的信息和能量等维度提取特征将脑电信号转换至区分性更强的特征域,最后利用SVM进行分类识别。采用国际BCI竞赛2003中的Graz数据进行验证,所提方法可以得到94.6%的正确识别率,为在线脑-机接口系统的研究提供了新的思路。     

Human emotion recognition using deep belief network architecture

Recently, deep learning methodologies have become popular to analyse physiological signals in multiple modalities via hierarchical architectures for human emotion recognition. In most of the state-of-the-arts of human emotion recognition, deep learning for emotion classification was used. However, deep learning is mostly effective for deep feature extraction. Therefore, in this research, we applied unsupervised deep belief network (DBN) for depth level feature extraction from fused observations of Electro-Dermal Activity (EDA), Photoplethysmogram (PPG) and Zygomaticus Electromyography (zEMG) sensors signals. Afterwards, the DBN produced features are combined with statistical features of EDA, PPG and zEMG to prepare a feature-fusion vector. The prepared feature vector is then used to classify five basic emotions namely Happy, Relaxed, Disgust, Sad and Neutral. As the emotion classes are not linearly separable from the feature-fusion vector, the Fine Gaussian Support Vector Machine (FGSVM) is used with radial basis function kernel for non-linear classification of human emotions. Our experiments on a public multimodal physiological signal dataset show that the DBN, and FGSVM based model significantly increases the accuracy of emotion recognition rate as compared to the existing state-of-the-art emotion classification techniques.     

DSP-based hierarchical neural network modulation signal classification

This paper discusses a real-time digital signal processor (DSP)-based hierarchical neural network classifier capable of classifying both analog and digital modulation signals. A high-performance DSP processor, namely the TMS320C6701, is utilized to implement different kinds of classifiers including a hierarchical neural network classifier. A total of 31 statistical signal features are extracted and used to classify 11 modulation signals plus white noise. The modulation signals include CW, AM, FM, SSB, FSK2, FSK4, PSK2, PSK4, OOK, QAM16, and QAM32. A classification hierarchy is introduced and the genetic algorithm is employed to obtain the most effective set of features at each level of the hierarchy. The classification results and the number of operations on the DSP processor indicate the effectiveness of the introduced hierarchical neural network classifier in terms of both classification rate and processing time.     

Classification of modulation signals using statistical signal characterization and artificial neural networks

Modulation recognition systems have to be able to correctly classify the incoming signal modulation scheme in the presence of noise. A new method for classification of analogue and digital modulated signals at low signal-to-noise ratio (SNR) is introduced in this paper. This method uses the statistical signal characterization (SSC) to extract parameters to classify the different modulation signals. The SSC technique produces a set of four numerical parameters for a specific modulated signal. Subsequent comparison of these parameters to those of other waveforms provides the basis for our classification system. The results of SSC technique are applied to an artificial neural network (ANN) to have a robust classification system in the presence of noise down to SNR of 3 dB. No a priori information is required by this technique about the set of input waveforms. The input to the classification system can be analogue or digital signals or a combination of both. The proposed technique shows a 100% efficiency of classification of analogue signals or digital signals at SNR of 7 dB. This classification efficiency reduces to 83% and 86% for analogue or digital signals at SNR of 3 dB. The SSC technique shows better classification results in comparison with other techniques with an important advantage over other methods, which is the simplicity of the neural network needed with this technique due to the small number of features used in the classification.     

Classification of EMG signals using combined features and soft computing techniques

The motor unit action potentials (MUPs) in an electromyographic (EMG) signal provide a significant source of information for the assessment of neuromuscular disorders. Since recently there were different types of developments in computer-aided EMG equipment, different methodologies in the time domain and frequency domain has been followed for quantitative analysis of EMG signals. In this study, the usefulness of the different feature extraction methods for describing MUP morphology is investigated. Besides, soft computing techniques were presented for the classification of intramuscular EMG signals. The proposed method automatically classifies the EMG signals into normal, neurogenic or myopathic. Also, multilayer perceptron neural networks (MLPNN), dynamic fuzzy neural network (DFNN) and adaptive neuro-fuzzy inference system (ANFIS) based classifiers were compared in relation to their accuracy in the classification of EMG signals. Concerning the impacts of features on the EMG signal classification, different results were obtained through analysis of the soft computing techniques. The comparative analysis suggests that the ANFIS modelling is superior to the DFNN and MLPNN in at least three points: slightly higher recognition rate; insensitivity to overtraining; and consistent outputs demonstrating higher reliability.