Heart sound classification based on scaled spectrogram and tensor decomposition

Heart sound signal analysis is an effective and convenient method for the preliminary diagnosis of heart disease. However, automatic heart sound classification is still a challenging problem which mainly reflected in heart sound segmentation and feature extraction from the corresponding segmentation results. In order to extract more discriminative features for heart sound classification, a scaled spectrogram and tensor decomposition based method was proposed in this study. In the proposed method, the spectrograms of the detected heart cycles are first scaled to a fixed size. Then a dimension reduction process of the scaled spectrograms is performed to extract the most discriminative features. During the dimension reduction process, the intrinsic structure of the scaled spectrograms, which contains important physiological and pathological information of the heart sound signals, is extracted using tensor decomposition method. As a result, the extracted features are more discriminative. Finally, the classification task is completed by support vector machine (SVM). Moreover, the proposed method is evaluated on three public datasets offered by the PASCAL classifying heart sounds challenge and 2016 PhysioNet challenge. The results show that the proposed method is competitive.     

基于深度卷积神经网络的心音分类算法

针对现有心音分类算法普适性差、依赖于对基本心音的精确分割、分类模型结构单一等问题,提出采用大量未经过精确分割的心音二维特征图训练深度卷积神经网络(CNN)的方法;首先采用滑动窗口方法和梅尔频率系数对心音信号进行预处理,得到大量未经过精确分割的心音特征图;然后利用深度CNN模型对心音特征图进行训练和测试;根据卷积层间连接方式的不同,设计了 3种深度CNN模型:基于单一连接的卷积神经网络、基于跳跃连接的卷积神经网络、基于密集连接的卷积神经网络;实验结果表明,基于密集连接的卷积神经网络比其他两种网络具备更大的潜力;与其他心音分类算法相比,该算法不依赖于对基本心音的精确分割,且在分类准确率、敏感性和特异性方面均有提升.     

Automatic phonocardiograph signal analysis for detecting heart valve disorders

Skilled cardiologists probe heart sounds by electronic stethoscope through human ears, but interpretations of heart sounds is a very special skill which is quite difficult to teach in a structured way. Because of this reason, automatic heart sound analysis in computer systems would be very helpful for medical staffs. This paper presents a complete heart sound analysis system covering from the segmentation of beat cycles to the final determination of heart conditions. The process of heart beat cycle segmentation includes autocorrelation for predicting the cycle time of a heart beat. The feature extraction pipeline includes stages of the short-time Fourier transform, the discrete cosine transform, and the adaptive feature selection. Many features are extracted, but only a few specific ones are selected for the classification of each hyperplane based on a systematic approach. The experiments are done by a public heart sound database released by Texas Heart Institute. A very promising recognition rate has been achieved.     

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。     

深度卷积神经网络在心音分类方法中的应用

通过分析心音信号对心脏早期的病理状态进行确诊具有重要的意义。提出了一种基于深度卷积神经网络的心音分类方法。将心音信号转化成具有时频特性的梅尔频谱系数（Mel Frequency Spectral Coefficient,MFSC）特征图,将其作为深度卷积神经网络模型的输入;利用深度卷积神经网络对MFSC特征图进行训练,引入中心损失函数建立最优的深度学习模型;测试阶段,先将心音信号转换成多张二维MFSC特征图,然后利用训练好的深度学习模型对其分类,最后利用多数表决原则判断心音信号的类别。针对人工标注的训练样本有限,导致模型训练正确率不高的问题,以心音的二维MFSC特征图为对象分别从时间域和频率域进行随机屏蔽处理进而扩充训练样本。实验结果表明,该方法在PASCAL心音数据集上进行测试,对正常、杂音、早搏三种心音的分类性能明显优于现有最好的方法。     

Classification of cardiac sound signals using constrained tunable-Q wavelet transform

The features extracted from the cardiac sound signals are commonly used for detection and identification of heart valve disorders. In this paper, we present a new method for classification of cardiac sound signals using constrained tunable-Q wavelet transform (TQWT). The proposed method begins with a constrained TQWT based segmentation of cardiac sound signals into heart beat cycles. The features obtained from heart beat cycles of separately reconstructed heart sounds and murmur can better represent the various types of cardiac sound signals than that from containing both. Therefore, heart sounds and murmur have been separated using constrained TQWT. Then the proposed novel raw feature set has been created by the parameters that have been optimized while constraining the output of TQWT together with that of extracted by using time-domain representation and Fourier–Bessel (FB) expansion of separated heart sounds and murmur. However, the adaptively selected features have been used to obtain the final feature set for subsequent classification of cardiac sound signals using least squares support vector machine (LS-SVM) with various kernel functions. The performance of the proposed method has been validated with publicly available datasets and the results have been compared with the existing short-time Fourier transform (STFT) based method. The proposed method shows higher percentage classification accuracy of 94.01 as compared to 93.53 of STFT based method. In comparison with STFT based method, it is noteworthy that the proposed method uses well defined and lower dimensionality of feature vector that can reduce the computational complexity.     

基于自适应阈值的心音分段方法仿真研究

针对心音信号频率低、易受干扰并含有大量杂音的特点,提出了一种心音分段新方法。首先,采用dB6小波进行5层小波分解做心音信号降噪处理;然后,采用了一种提取降噪后心音信号连续平均能量包络的方法;最后,根据连续平均能量包络及自相关函数提出了一种自适应阈值心音分段方法。仿真结果表明,该方法所提取的心音信号包络特征更稳健,提出的分段算法与基于短时能熵比法和短时自相关函数法等心音分段算法相比,本文所提算法分段准确度更高。     

基于CNN和RNN联合网络的心音自动分类

为充分利用心音的全局信息,提出不依赖于分割的心音自动分类方法。对目前的心音分类方法进行总结,分析单阶段和两阶段方法的优势与不足,提出以深度学习提取更好的全局特征作为提升分类效果的新方向。使用精调的卷积神经网络和循环神经网络分别提取心音的频域和时域特征,辅以数据增强的方法进行训练。该方法在测试集的平均分类准确率达到了85.7%,达到了目前单阶段心音分类方法中的最好效果。     

Spectral–spatial classification of hyperspectral images by algebraic multigrid based multiscale information fusion

In this work, we present a novel spectral-spatial classification framework of hyperspectral images (HSIs) by integrating the techniques of algebraic multigrid (AMG), hierarchical segmentation (HSEG) and Markov random field (MRF). The proposed framework manifests two main contributions. First, an effective HSI segmentation method is developed by combining the AMG-based marker selection approach and the conventional HSEG algorithm to construct a set of unsupervised segmentation maps in multiple scales. To improve the computational efficiency, the fast Fish Markov selector (FMS) algorithm is exploited for feature selection before image segmentation. Second, an improved MRF energy function is proposed for multiscale information fusion (MIF) by considering both spatial and inter-scale contextual information. Experiments were performed using two airborne HSIs to evaluate the performance of the proposed framework in comparison with several popular classification methods. The experimental results demonstrated that the proposed framework can provide superior performance in terms of both qualitative and quantitative analysis.     

A Unified Neural Network for Panoptic Segmentation

In this paper, we propose a unified neural network for panoptic segmentation, a task aiming to achieve more fine‐grained segmentation. Following existing methods combining semantic and instance segmentation, our method relies on a triple‐branch neural network for tackling the unifying work. In the first stage, we adopt a ResNet50 with a feature pyramid network (FPN) as shared backbone to extract features. Then each branch leverages the shared feature maps and serves as the stuff, things, or mask branch. Lastly, the outputs are fused following a well‐designed strategy. Extensive experimental results on MS‐COCO dataset demonstrate that our approach achieves a competitive Panoptic Quality (PQ) metric score with the state of the art.     

基于深度感知特征提取的室内场景理解

从深度图RGB-D域中联合学习RGB图像特征与3D几何信息有利于室内场景语义分割,然而传统分割方法通常需要精确的深度图作为输入,严重限制了其应用范围。提出一种新的室内场景理解网络框架,建立基于语义特征与深度特征提取网络的联合学习网络模型提取深度感知特征,通过几何信息指导的深度特征传输模块与金字塔特征融合模块将学习到的深度特征、多尺度空间信息与语义特征相结合,生成具有更强表达能力的特征表示,实现更准确的室内场景语义分割。实验结果表明,联合学习网络模型在NYU-Dv2与SUN RGBD数据集上分别取得了69.5%与68.4%的平均分割准确度,相比传统分割方法具有更好的室内场景语义分割性能及更强的适用性。     

Adaptive neuro-fuzzy inference system for diagnosis of the heart valve diseases using wavelet transform with entropy

Listening via stethoscope is a preferential method, being used by physicians for distinguishing normal and abnormal cardiac systems. On the other hand, listening with stethoscope has a number of constraints. The interpretation of various heart sounds depends on physician’s ability of hearing, experience, and skill. Such limitations may be reduced by developing biomedical-based decision support systems. In this study, a biomedical-based decision support system was developed for the classification of heart sound signals, obtained from 120 subjects with normal, pulmonary, and mitral stenosis heart valve diseases via stethoscope. Developed system comprises of three stages. In the first stage, for feature extraction, obtained heart sound signals were separated to its sub-bands using discrete wavelet transform (DWT). In the second stage, entropy of each sub-band was calculated using Shannon entropy algorithm to reduce the dimensionality of the feature vectors via DWT. In the third stage, the reduced features of three types of heart sound signals were used as input patterns of the adaptive neuro-fuzzy inference system (ANFIS) classifiers. Developed method reached 98.33% classification accuracy, and it was showed that purposed method is effective for detection of heart valve diseases.     

基于神经网络的遥感图像海陆语义分割方法

针对海陆语义分割中陆地、码头形状多样,背景目标复杂等情况造成的像素分类错误、边界分割模糊等问题,提出了一种新的基于深度卷积神经网络的遥感图像海陆语义分割方法。该方法以端对端的训练方式实现了对目标的逐像素分类,为了解决海陆分割中像素分类错误,设计以不同尺度图像为输入的三个并行的编码结构,通过融合不同尺度的特征图,丰富特征代表算子的语义信息,增大像素分类准确率。为了解决海陆分割中边界分割模糊,通过设计能够融合编码结构中低层精细位置信息的解码结构,对特征图进行更加精确的上采样,恢复像素的密集位置信息,提高海陆分割准确度。为有效验证所提网络框架的优势,构建了海陆分割数据集HRSC2016-SL进行算法性能比较。与最新的语义分割算法相比,所提算法取得了更好的分割结果。     

Wavelet ridges for musical instrument classification

The time-varying frequency structure of musical signals have been analyzed using wavelets by either extracting the instantaneous frequency of signals or building features from the energies of sub-band coefficients. We propose to benefit from a combination of these two approaches and use the time-frequency domain energy localization curves, called as wavelet ridges, in order to build features for classification of musical instrument sounds. We evaluated the representative capability of our feature in different musical instrument classification problems using support vector machine classifiers. The comparison with the features based on parameterizing the wavelet sub-band energies confirmed the effectiveness of the proposed feature.     

结合上下文特征与CNN多层特征融合的语义分割

目的 针对基于区域的语义分割方法在进行语义分割时容易缺失细节信息,造成图像语义分割结果粗糙、准确度低的问题,提出结合上下文特征与卷积神经网络（CNN）多层特征融合的语义分割方法。方法 首先,采用选择搜索方法从图像中生成不同尺度的候选区域,得到区域特征掩膜;其次,采用卷积神经网络提取每个区域的特征,并行融合高层特征与低层特征。由于不同层提取的特征图大小不同,采用RefineNet模型将不同分辨率的特征图进行融合;最后将区域特征掩膜和融合后的特征图输入到自由形式感兴趣区域池化层,经过softmax分类层得到图像的像素级分类标签。结果 采用上下文特征与CNN多层特征融合作为算法的基本框架,得到了较好的性能,实验内容主要包括CNN多层特征融合、结合背景信息和融合特征以及dropout值对实验结果的影响分析,在Siftflow数据集上进行测试,像素准确率达到82.3%,平均准确率达到63.1%。与当前基于区域的端到端语义分割模型相比,像素准确率提高了10.6%,平均准确率提高了0.6%。结论 本文算法结合了区域的前景信息和上下文信息,充分利用了区域的语境信息,采用弃权原则降低网络的参数量,避免过拟合,同时利用RefineNet网络模型对CNN多层特征进行融合,有效地将图像的多层细节信息用于分割,增强了模型对于区域中小目标物体的判别能力,对于有遮挡和复杂背景的图像表现出较好的分割效果。     

基于对象位置线索的弱监督图像语义分割方法

深度卷积神经网络使用像素级标注,在图像语义分割任务中取得了优异的分割性能.然而,获取像素级标注是一项耗时并且代价高的工作.为了解决这个问题,提出一种基于图像级标注的弱监督图像语义分割方法.该方法致力于使用图像级标注获取有效的伪像素标注来优化分割网络的参数.该方法分为3个步骤：（1）首先,基于分类与分割共享的网络结构,通过空间类别得分（图像二维空间上像素点的类别得分）对网络特征层求导,获取具有类别信息的注意力图;（2）采用逐次擦除法产生显著图,用于补充注意力图中缺失的对象位置信息;（3）融合注意力图与显著图来生成伪像素标注并训练分割网络.在PASCAL VOC 2012分割数据集上的一系列对比实验,证明了该方法的有效性及其优秀的分割性能.     

Image segmentation scheme based on SOM–PCNN in frequency domain

A hybrid scheme for the image segmentation of high-resolution images is proposed in this study. Our methodology is based on combining both supervised and unsupervised segmentation. The entire process is performed in the frequency domain, rather than the spatial domain, using the Shift Invariant Shearlet Transform (SIST). Initially, the input image is filtered using an anisotropic filter to enhance the texture features. Then, it is separated into low and high sub-band frequencies using SIST. Subsequently, we built a feature vector from coarser coefficients complemented with texture information extracted from high-frequency coefficients of the input image. SOM is used for the preliminary classification of the input image coefficients, and the network training process is performed using the previously built feature vector. Lastly, the modified PCNN is used to augment the SOM results to reduce the over-segmentation artefacts. We used the Berkeley Segmentation Database (BSR) and Quick-Bird Satellite images to validate the results. It was found that the proposed scheme is superior to the Fuzzy-C-Means-based, SOM-based, and PCNN-based segmentation algorithms in terms of quantitative criteria and visual interpretation.     

基于PCA串并融合的多路心音特征表征方法

多路心音信号不仅比单路心音信号涵盖更多关于总体的特征,而且能够弥补单路心音数据携带的信息量可能不充分的缺陷。利用笔者自主设计的4路心音传感器,初步建立一个小型4路心音数据库。基于这个数据库,首先阐明多路心音信号的特点,论述心杂音与听诊位置的关系;然后分别提取心音的单路和4路能量熵系数、4路心音互信息作为有效特征数据集,利用PCA对能量熵特征进行降维处理,获得串行特征;将相关性特征和互信息特征从实向量空间拓展到复向量空间,进行并行融合,获得并行特征;最后将串行并行特征再次融合成为多元优化组合特征。这种融合策略,具有针对性强,凸显差异性的优点。仿真实验结果表明,由多路心音信号获取的多元优化组合特征表征效果明显优于单路心音信号的特征表征效果,不仅有益于分类模型的构建,而且对实现先心病的快速筛查,提高分类识别率具有积极的意义。     

小波和神经网络在心音识别中的应用

提出一种心音的特征提取和分类方法,用离散小波变换分解、重构产生信号的细节包络,进而用于提取特征,从预处理的信号中提取统计特性,作为心音分类的特征。多层感知器用于心音的分类,并通过250个心动周期得到验证,算法识别率达到92%。