基于功率谱密度与卷积神经网络的心音分类

正常与异常心音分类在心血管疾病的筛查中有着重要的作用。建立在无心音分割的基础上,提出了一种基于功率谱密度时频分布特征与卷积神经网络的心音分类方法。该方法采用小波降噪做预处理,通过循环自相关获取心动周期,采用双线性插值法提取维度一致的心动周期功率谱密度时频特征,并送入卷积神经网络进行训练与测试。实验采用Challenge 2016数据集进行训练与测试,测试集的分类精度达到0.847 2,灵敏度和特异性评分达到0.776 3和0.946 3,整体性能良好。与其他算法的对比结果显示,该算法获得了更高的总体评分。     

Classification of intelligence quotient via brainwave sub-band power ratio features and artificial neural network

This paper elaborates on the novel intelligence assessment method using the brainwave sub-band power ratio features. The study focuses only on the left hemisphere brainwave in its relaxed state. Distinct intelligence quotient groups have been established earlier from the score of the Raven Progressive Matrices. Sub-band power ratios are calculated from energy spectral density of theta, alpha and beta frequency bands. Synthetic data have been generated to increase dataset from 50 to 120. The features are used as input to the artificial neural network. Subsequently, the brain behaviour model has been developed using an artificial neural network that is trained with optimized learning rate, momentum constant and hidden nodes. Findings indicate that the distinct intelligence quotient groups can be classified from the brainwave sub-band power ratios with 100% training and 88.89% testing accuracies.     

Modified mixture of experts employing eigenvector methods and Lyapunov exponents for analysis of electroencephalogram signals

Abstract: The use of diverse features in detecting variability of electroencephalogram (EEG) signals is presented. The classification accuracies of the modified mixture of experts (MME), which was trained on diverse features, were obtained. Eigenvector methods (Pisarenko, multiple signal classification – MUSIC, and minimum-norm) were selected to generate the power spectral density estimates. The features from the power spectral density estimates and Lyapunov exponents of the EEG signals were computed and statistical features were calculated to depict their distribution. The statistical features, which were used for obtaining the diverse features of the EEG signals, were then input into the implemented neural network models for training and testing purposes. The present study demonstrated that the MME trained on the diverse features achieved high accuracy rates (total classification accuracy of the MME is 98.33%).     

VLSI implementation of transcendental function hyperbolic tangent for deep neural network accelerators

Extensive use of neural network applications prompted researchers to customize a design to speed up their computation based on ASIC implementation. The choice of activation function (AF) in a neural network is an essential requirement. Accurate design architecture of an AF in a digital network faces various challenges as these AF require more hardware resources because of its non-linear nature. This paper proposed an efficient approximation scheme for hyperbolic tangent (tanh) function which purely based on combinational design architecture. The approximation is based on mathematical analysis by considering maximum allowable error in a neural network. The results prove that the proposed combinational design of an AF is efficient in terms of area, power and delay with negligible accuracy loss on MNIST and CIFAR-10 benchmark datasets. Post synthesis results show that the proposed design area is reduced by 66% and delay is reduced by nearly 16% compared to state-of-the-art.     

ECG beat classifier designed by combined neural network model

This paper illustrates the use of combined neural network model to guide model selection for classification of electrocardiogram (ECG) beats. The ECG signals were decomposed into time-frequency representations using discrete wavelet transform and statistical features were calculated to depict their distribution. The first level networks were implemented for ECG beats classification using the statistical features as inputs. To improve diagnostic accuracy, the second level networks were trained using the outputs of the first level networks as input data. Four types of ECG beats (normal beat, congestive heart failure beat, ventricular tachyarrhythmia beat, atrial fibrillation beat) obtained from the Physiobank database were classified with the accuracy of 96.94% by the combined neural network. The combined neural network model achieved accuracy rates which were higher than that of the stand-alone neural network model.     

Statistics over features of ECG signals

This paper presented the usage of statistics over the set of the features representing the electrocardiogram (ECG) signals. Since classification is more accurate when the pattern is simplified through representation by important features, feature extraction and selection play an important role in classifying systems such as neural networks. Multilayer perceptron neural network (MLPNN) architectures were formulated and used as basis for detection of variabilities of the ECG signals. Four types of ECG beats (normal beat, congestive heart failure beat, ventricular tachyarrhythmia beat, atrial fibrillation beat) obtained from the Physiobank database were classified. The selected Lyapunov exponents, wavelet coefficients and the power levels of power spectral density (PSD) values obtained by eigenvector methods of the ECG signals were used as inputs of the MLPNN trained with Levenberg–Marquardt algorithm. The classification results confirmed that the proposed MLPNN has potential in detecting the variabilities of the ECG signals.     

Homologue matching applications: Recognition of overlapped chromosomes

Automated Giemsa-banded chromosome image research has been largely restricted to classification schemes associated with isolated chromosomes within metaphase spreads. Overlapping chromosomes cause difficulties in the automated chromosome karyotyping process. First, overlapping chromosomes must be recognised and decomposed into the proper chromosome parts. Secondly, the decomposed chromosomes must be classified. The first difficulty is associated with image segmentation. The second area is a pattern recognition problem. Even if chromosomes within overlapping clusters are decomposed properly, classification capability is impaired due to feature distortion in the overlapped regions. In normal human metaphase spreads, chromosomes occur in homologous pairs for the autosomal classes, 1–22, and X chromosome for females. This research presents a homologue matching approach for overlapped chromosome recognition. The undistorted grey level information in isolated chromosomes is used for identifying overlapped chromosomes. An isolated chromosome prototype is obtained using neural networks. Dynamic programming and neural networks are compared for matching the prototype to its overlapped homoloque. The homologue matching method is applied to identifying chromosome 2 in 50 metaphase spreads. Experimental results showed that homologue matching using dynamic programming matching based on the density profile achieved a higher correct recognition rate than homologue matching using three different neural network approaches.Grant Support: This research was supported in part by a grant from the University of Missouri Research board and training grant 5 T15 LM/CA07089-04 from the National Library of Medicine and National Cancer Institute, Bethesda, MD.     

Structure damage diagnosis using neural network and feature fusion

A structure damage diagnosis method combining the wavelet packet decomposition, multi-sensor feature fusion theory and neural network pattern classification was presented. Firstly, vibration signals gathered from sensors were decomposed using orthogonal wavelet. Secondly, the relative energy of decomposed frequency band was calculated. Thirdly, the input feature vectors of neural network classifier were built by fusing wavelet packet relative energy distribution of these sensors. Finally, with the trained classifier, damage diagnosis and assessment was realized. The result indicates that, a much more precise and reliable diagnosis information is obtained and the diagnosis accuracy is improved as well.     

基于模糊神经网络的二自由度主动悬架滑模控制系统设计

主动悬架滑模控制系统受到行驶路面影响,车体震动加速度功率谱密度与实际密度不一致,导致系统控制效果不佳,提出基于模糊神经网络的二自由度主动悬架滑模控制系统设计;基于二自由度硬件结构,在主动悬架液压伺服系统中安装蓄能器,减小空间内存;采用磁流变阻尼器,调整电流大小控制阻尼;设计自适应减振座椅悬架并计算阻尼值,实现了主动悬架的控制;考虑路面不规则度,构建二自由度主动悬架滑模控制模型,依据牛顿定律计算悬架弹性元件受力;采用模糊控制规则校正控制误差,采用模糊神经网络设计主动悬架滑膜控制方案,经迭代处理得到满足优化要求的解,实现滑膜控制;由试验结果可知,车体震动加速度功率谱密度与实际密度一致,从最初的0 ms^(-2)到最终的1.5 ms^(-2),具有良好控制效果,确保乘车的舒适性。     

EEG signal classification using wavelet feature extraction and a mixture of expert model

Mixture of experts (ME) is modular neural network architecture for supervised learning. A double-loop Expectation-Maximization (EM) algorithm has been introduced to the ME network structure for detection of epileptic seizure. The detection of epileptiform discharges in the EEG is an important component in the diagnosis of epilepsy. EEG signals were decomposed into the frequency sub-bands using discrete wavelet transform (DWT). Then these sub-band frequencies were used as an input to a ME network with two discrete outputs: normal and epileptic. In order to improve accuracy, the outputs of expert networks were combined according to a set of local weights called the “gating function”. The invariant transformations of the ME probability density functions include the permutations of the expert labels and the translations of the parameters in the gating functions. The performance of the proposed model was evaluated in terms of classification accuracies and the results confirmed that the proposed ME network structure has some potential in detecting epileptic seizures. The ME network structure achieved accuracy rates which were higher than that of the stand-alone neural network model.     

Implementing wavelet transform/mixture of experts network for analysis of electrocardiogram beats

Abstract: Mixture of experts (ME) is a modular neural network architecture for supervised learning. This paper illustrates the use of the ME network structure to guide model selection for classification of electrocardiogram (ECG) beats. The expectation maximization algorithm is used for training the ME so that the learning process is decoupled in a manner that fits well with the modular structure. The ECG signals were decomposed into time–frequency representations using discrete wavelet transforms and statistical features were calculated to depict their distribution. The ME network structure was implemented for ECG beats classification using the statistical features as inputs. To improve classification accuracy, the outputs of expert networks were combined by a gating network simultaneously trained in order to stochastically select the expert that is performing the best at solving the problem. Five types of ECG beats (normal beat, congestive heart failure beat, ventricular tachyarrhythmia beat, atrial fibrillation beat, partial epilepsy beat) obtained from the Physiobank database were classified with an accuracy of 96.89% by the ME network structure. The ME network structure achieved accuracy rates which were higher than those of the stand-alone neural network models.     

Automatic identification of cardiac health using modeling techniques: A comparative study

Heart rate variability (HRV), a widely adopted quantitative marker of the autonomic nervous system can be used as a predictor of risk of cardiovascular diseases. Moreover, decreased heart rate variability (HRV) has been associated with an increased risk of cardiovascular diseases. Hence in this work HRV signal is used as the base signal for predicting the risk of cardiovascular diseases. The present study concerns nine cardiac classes that include normal sinus rhythm (NSR), congestive heart failure (CHF), atrial fibrillation (AF), ventricular fibrillation (VF), preventricular contraction (PVC), left bundle branch block (LBBB), complete heart block (CHB), ischemic/dilated cardiomyopathy (ISCH) and sick sinus syndrome (SSS). A total of 352 cardiac subjects belonging to the nine classes were analyzed in the frequency domain. The fast Fourier transforms (FFT) and three other modeling techniques namely, autoregressive (AR) model, moving average (MA) model and the autoregressive moving average (ARMA) model are used to estimate the power spectral densities of the RR interval variability. The spectral parameters obtained from the spectral analysis of the HRV signals are used as the input parameters to the artificial neural network (ANN) for classification of the different cardiac classes. Our findings reveal that the ARMA modeling technique seems to give better resolution and would be more promising for clinical diagnosis.     

Combined neural network model employing wavelet coefficients for EEG signals classification

This paper illustrates the use of combined neural network model to guide model selection for classification of electroencephalogram (EEG) signals. The EEG signals were decomposed into time–frequency representations using discrete wavelet transform and statistical features were calculated to depict their distribution. The first-level networks were implemented for the EEG signals classification using the statistical features as inputs. To improve diagnostic accuracy, the second-level networks were trained using the outputs of the first-level networks as input data. Three types of EEG signals (EEG signals recorded from healthy volunteers with eyes open, epilepsy patients in the epileptogenic zone during a seizure-free interval, and epilepsy patients during epileptic seizures) were classified with the accuracy of 94.83% by the combined neural network. The combined neural network model achieved accuracy rates which were higher than that of the stand-alone neural network model.     

Signal recognition: Both components of the short time Fourier transform vs. power spectral density

A new feature representation approach was generalised and used for Gaussian recognition. The generalised approach consists of simultaneously using two new recognition features — real and imaginary Fourier components —taking into account the covariance between features. Generalisation of the approach improves recognition effectiveness. An advanced time-frequency technique, the short time Fourier transform, was considered. Covariance and the correlation coefficient between the proposed features were obtained for the first time for arbitrary stationary signals. The recognition effectiveness between the generalised approach and power spectral density was compared. It was shown that power spectral density is not an optimal feature, and represents only a particular case of the generalised approach. The use of power spectral density is optimal if simultaneously the correlation coefficient between Fourier components is equal to zero, and the standard deviations of components are equal. Use of the generalised approach provides an increase in effectiveness in comparison with power spectral density.     

基于改进混合粒子群算法和匹配理论的无人机电力巡检卸载策略

无人机搭载深度神经网络进行自主电力巡检时由于受到设备本身计算能力、电池容量、深度神经网络计算负载的限制,无法独立处理巡检任务中产生的海量图像数据。为解决该问题,提出了一种基于改进混合粒子群算法和匹配理论的无人机电力巡检卸载策略,该策略将系统成本最小化问题分解为深度神经网络计算任务协同分割和边缘服务器选择两个子问题。针对协同分割子问题,基于深度神经网络计算任务的执行流程提出了一种错时传输方法,通过改进混合粒子群算法求解多无人机任务协同分割层。针对边缘服务器选择子问题,定义无人机与边缘服务器各自偏好函数,根据偏好函数通过匹配理论建立两者间的稳定匹配,得到边缘服务器选择策略。仿真结果表明,与其他卸载策略相比,所提策略能有效降低无人机能耗和计算任务处理时延,促进边缘服务器负载均衡。     

Wavelet energy signatures and robust Bayesian neural network for visual quality recognition of nonwovens

In this paper, the visual quality recognition of nonwovens is considered as a common problem of pattern recognition that will be solved by a joint approach by combining wavelet energy signatures, Bayesian neural network, and outlier detection. In this research, 625 nonwovens images of 5 different grades, 125 each grade, are decomposed at 4 levels with wavelet base sym6, then two energy signatures, norm-1 L¹ and norm-2 L² are calculated from wavelet coefficients of each high frequency subband to train and test Bayesian neural network. To detect the outlier of training set, scaled outlier probability of training set and outlier probability of each sample are introduced. The committees of networks and the evidence criterion are employed to select the ‘most suitable’ model, given a set of candidate networks which has different numbers of hidden neurons. However, in our research with the finite industrial data, we take both the evidence criterion and the actual performance into account to determine the structure of Bayesian neural network. When the nonwoven images are decomposed at level 4, with 500 samples to training the Bayesian neural network that has 3 hidden neurons, the average recognition accuracy of test set is 99.2%. Experimental results on the 625 nonwoven images indicate that the wavelet energy signatures are expressive and powerful in characterizing texture of nonwoven images and the robust Bayesian neural network has excellent recognition performance.     

基于脑电时频空多域特征融合的情感识别研究

传统基于脑电信号（electroencephalogram,EEG）的情感识别主要采用单一的脑电特征提取方法,为了充分利用EEG中蕴含的丰富信息,提出一种多域特征融合的脑电情感识别新方法。提取了EEG的时域、频域和空域特征,将三域特征进行融合作为情感识别模型的输入。首先计算不同时间窗EEG信号的alpha、beta、gamma三个频段功率谱密度,并结合脑电电极空间信息构成EEG图片,然后利用卷积神经网络（convolutional neural network,CNN）与双向长短期记忆网络（bidirectional long short-term memory network,BLSTM）构建CNN-BLSTM情感识别模型,分别对时、频、空三域特征进行学习。在SEED数据集对该方法进行验证,结果表明该方法能有效提高情感识别精度,平均识别准确率达96.25%。     

一种网络流量预测的小波神经网络模型

结合小波变换和人工神经网络的优势,建立一种网络流量预测的小波神经网络模型。首先对流量时间序列进行小波分解,得到小波变换尺度系数序列和小波系数序列,以系数序列和原来的流量时间序列分别作为模型的输入和输出,构造人工神经网络并且加以训练。用实际网络流量对该模型进行验证,结果表明,该模型具有较高的预测效果。     

EEG based automated detection of auditory loss: A pilot study

Auditory dysfunction is one of the most common deficiencies present in the newborn. Recent studies show that significant bilateral hearing loss is present in ∼1 to 3 per 1000 newborn infants in the well-baby nursery population and in ∼2 to 4 per 1000 infants in the ICU population. The ignorance of hearing screening test at the initial stage will impede speech, language and cognitive development. It has been further noted that direct physician observation as well as parental recognition has not been significantly successful until today in identifying the hearing loss in the first year of life. To overcome such problems, early screening is essential. This paper presents a pilot study on detection of hearing loss by applying electroencephalography (EEG) signals as the key indicator. The effect of auditory evoked potential (AEP) is exploited on EEGs by introducing an external stimulus to the subject’s auditory canal. Two time domain features, spike rhythmicity, autoregressive model using Levinson-Durbin algorithm and frequency domain features such as power spectral density estimation by Burg’s and Yule-Walker methods are applied. Feed forward and feedback neural network models are used to distinguish the stimuli and non-stimuli EEGs. The neural network models are configured optimally by varying the hidden neurons and learning algorithms and their performance are evaluated in terms of specificity, sensitivity and classification accuracy. It can be concluded from the experimental study that the proposed methodology can be applied for neonatal healthcare applications.     

Monitoring and control of power system voltage stability using an artificial neural network

This paper proposes a decentralized method for monitoring and control of power system voltage stability with an artificial neural network (ANN). One of the problems in applying neural networks to power systems is how to cope with a lot of variables in real-size power systems. Most of the conventional ANN-based approaches suffer from the curse of the dimensionality in power systems. As a result, it seems that the applications are far from the real world. However, as far as voltage problems are concerned, they possess peculiar local characteristics. It implies that the problem may be decomposed into subproblems. This paper focuses on the characteristics and considers more realistic ANN applications. The proposed method is tested in a 30-node system to demonstrate the effectiveness.