Pub Date : 2021-01-24DOI: 10.23919/Eusipco47968.2020.9287557
K. Chumachenko, Alexandros Iosifidis, M. Gabbouj
In this paper, we propose novel methods to address the challenges of dimensionality reduction related to potential outlier classes and imbalanced classes often present in data. In particular, we propose extensions to Fast Subclass Discriminant Analysis and Subclass Discriminant Analysis that allow to put more attention on uder-represented classes or classes that are likely to be confused with each other. Furthermore, the kernelized variants of the proposed algorithms are presented. The proposed methods lead to faster training time and improved accuracy as shown by experiments on eight datasets of different domains, tasks, and sizes.
{"title":"Robust Fast Subclass Discriminant Analysis","authors":"K. Chumachenko, Alexandros Iosifidis, M. Gabbouj","doi":"10.23919/Eusipco47968.2020.9287557","DOIUrl":"https://doi.org/10.23919/Eusipco47968.2020.9287557","url":null,"abstract":"In this paper, we propose novel methods to address the challenges of dimensionality reduction related to potential outlier classes and imbalanced classes often present in data. In particular, we propose extensions to Fast Subclass Discriminant Analysis and Subclass Discriminant Analysis that allow to put more attention on uder-represented classes or classes that are likely to be confused with each other. Furthermore, the kernelized variants of the proposed algorithms are presented. The proposed methods lead to faster training time and improved accuracy as shown by experiments on eight datasets of different domains, tasks, and sizes.","PeriodicalId":6705,"journal":{"name":"2020 28th European Signal Processing Conference (EUSIPCO)","volume":"51 1","pages":"1397-1401"},"PeriodicalIF":0.0,"publicationDate":"2021-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73788421","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-24DOI: 10.23919/Eusipco47968.2020.9287679
Berkan Kiliç, A. Güngör, M. Kalfa, O. Arikan
Design of measurement matrices is an important aspect of compressed sensing (CS) based direction of arrival (DoA) applications that enables reduction in the analog channels to be processed in sparse target environments. Here, a novel measurement matrix design methodology for CS based DoA estimation is proposed and its superior performance over alternative measurement matrix design methodologies is demonstrated. The proposed method uses prior probability distribution of the targets to improve performance. Compared to the state-of-the-art techniques, it is quantitatively demonstrated that the proposed measurement matrix design approach enables significant reduction in the number of analog channels to be processed and adapts to a priori information on the target scene.
{"title":"Adaptive Measurement Matrix Design in Compressed Sensing Based Direction of Arrival Estimation","authors":"Berkan Kiliç, A. Güngör, M. Kalfa, O. Arikan","doi":"10.23919/Eusipco47968.2020.9287679","DOIUrl":"https://doi.org/10.23919/Eusipco47968.2020.9287679","url":null,"abstract":"Design of measurement matrices is an important aspect of compressed sensing (CS) based direction of arrival (DoA) applications that enables reduction in the analog channels to be processed in sparse target environments. Here, a novel measurement matrix design methodology for CS based DoA estimation is proposed and its superior performance over alternative measurement matrix design methodologies is demonstrated. The proposed method uses prior probability distribution of the targets to improve performance. Compared to the state-of-the-art techniques, it is quantitatively demonstrated that the proposed measurement matrix design approach enables significant reduction in the number of analog channels to be processed and adapts to a priori information on the target scene.","PeriodicalId":6705,"journal":{"name":"2020 28th European Signal Processing Conference (EUSIPCO)","volume":"58 1","pages":"1881-1885"},"PeriodicalIF":0.0,"publicationDate":"2021-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73102113","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-24DOI: 10.23919/Eusipco47968.2020.9287759
Cédrick Bamba Nsimba, A. Levada
This paper applies a two-step approach for texture classification by combining Manifold learning with Deep CNN feature extractors. The first step is to use CNN architecture to compute the feature vector of a given image. The second step is to apply Manifold Learning algorithms on the features computed in the first step to making a refined feature vector. Eventually, this final representation is used to train SVM classifier. In the first step, we adopted VGG-19 network trained from scratch in order to extract texture features. In the next step, we used the DIMAL (Deep Isometric Manifold Learning Using Sparse Geodesic Sampling) configuration to train a neural network to reduce the dimensionality of the feature space in a nonlinear manner for generating the refined feature vector of the input image. Our concept is that the combination of a deep-learning framework with manifold learning techniques has the potential to select discriminative texture features from a high dimensional space. Based on this idea, we adopted this combination to perform nonlinear feature extraction in texture images. The resulting learned features were then used to train SVM classifier. The experiments demonstrated that our approach achieved better accuracy in texture classification than existing models if trained from scratch.
本文将流形学习与深度CNN特征提取器相结合，采用两步法进行纹理分类。第一步是使用CNN架构计算给定图像的特征向量。第二步是将流形学习算法应用到第一步计算的特征上，得到一个精细的特征向量。最后，将此最终表示用于训练SVM分类器。第一步，我们采用从头开始训练的VGG-19网络提取纹理特征。在下一步中，我们使用DIMAL (Deep Isometric Manifold Learning Using Sparse Geodesic Sampling)配置来训练神经网络，以非线性方式降低特征空间的维数，以生成输入图像的精细特征向量。我们的概念是，深度学习框架与流形学习技术的结合有可能从高维空间中选择有区别的纹理特征。基于这一思想，我们采用这种组合对纹理图像进行非线性特征提取。然后使用学习得到的特征来训练SVM分类器。实验表明，我们的方法在纹理分类上取得了比现有模型更好的准确率。
{"title":"Combining Deep and Manifold Learning For Nonlinear Feature Extraction in Texture Images","authors":"Cédrick Bamba Nsimba, A. Levada","doi":"10.23919/Eusipco47968.2020.9287759","DOIUrl":"https://doi.org/10.23919/Eusipco47968.2020.9287759","url":null,"abstract":"This paper applies a two-step approach for texture classification by combining Manifold learning with Deep CNN feature extractors. The first step is to use CNN architecture to compute the feature vector of a given image. The second step is to apply Manifold Learning algorithms on the features computed in the first step to making a refined feature vector. Eventually, this final representation is used to train SVM classifier. In the first step, we adopted VGG-19 network trained from scratch in order to extract texture features. In the next step, we used the DIMAL (Deep Isometric Manifold Learning Using Sparse Geodesic Sampling) configuration to train a neural network to reduce the dimensionality of the feature space in a nonlinear manner for generating the refined feature vector of the input image. Our concept is that the combination of a deep-learning framework with manifold learning techniques has the potential to select discriminative texture features from a high dimensional space. Based on this idea, we adopted this combination to perform nonlinear feature extraction in texture images. The resulting learned features were then used to train SVM classifier. The experiments demonstrated that our approach achieved better accuracy in texture classification than existing models if trained from scratch.","PeriodicalId":6705,"journal":{"name":"2020 28th European Signal Processing Conference (EUSIPCO)","volume":"102 1","pages":"1552-1555"},"PeriodicalIF":0.0,"publicationDate":"2021-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75749530","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-24DOI: 10.23919/Eusipco47968.2020.9287833
A. Gkillas, D. Ampeliotis, K. Berberidis
The problem of computing a proper sparse representation matrix for a signal matrix that obeys some local smoothness property, given an over-complete dictionary, is considered. The focus is on piece-wise smooth signals, defined as signals that comprise a number of blocks that each fulfills the considered smoothness property. A computationally efficient sparse coding algorithm is derived by limiting the number of times that a new support set of dictionary atoms is computed, exploiting the smoothness of the signal. Furthermore, a new, total-variation regularized problem is proposed for computing the required sparse coding coefficients, exploiting further the smoothness priors of the signals. The considered problem is solved using the alternating direction method of multipliers. Finally, numerical results considering hyperspectral images are provided, that demonstrate the applicability and complexity -denoising performance benefits of the novel algorithms.
{"title":"Fast Sparse Coding Algorithms for Piece-wise Smooth Signals","authors":"A. Gkillas, D. Ampeliotis, K. Berberidis","doi":"10.23919/Eusipco47968.2020.9287833","DOIUrl":"https://doi.org/10.23919/Eusipco47968.2020.9287833","url":null,"abstract":"The problem of computing a proper sparse representation matrix for a signal matrix that obeys some local smoothness property, given an over-complete dictionary, is considered. The focus is on piece-wise smooth signals, defined as signals that comprise a number of blocks that each fulfills the considered smoothness property. A computationally efficient sparse coding algorithm is derived by limiting the number of times that a new support set of dictionary atoms is computed, exploiting the smoothness of the signal. Furthermore, a new, total-variation regularized problem is proposed for computing the required sparse coding coefficients, exploiting further the smoothness priors of the signals. The considered problem is solved using the alternating direction method of multipliers. Finally, numerical results considering hyperspectral images are provided, that demonstrate the applicability and complexity -denoising performance benefits of the novel algorithms.","PeriodicalId":6705,"journal":{"name":"2020 28th European Signal Processing Conference (EUSIPCO)","volume":"35 1","pages":"2040-2044"},"PeriodicalIF":0.0,"publicationDate":"2021-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79779125","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-24DOI: 10.23919/Eusipco47968.2020.9287421
Giovanni Pepe, L. Gabrielli, S. Squartini, L. Cattani, Carlo Tripodi
In this paper we present an evolutionary algorithm for the design of stable IIR filters for binaural audio equalization. The filters are arranged as a cascade of second-order sections (SOS’s) and the gravitational search algorithm (GSA) is used. This process seeks for optimal coefficients based on a fitness function, possibly leading to unstable filters. To avoid this, we propose two alternative methods. Experiments have been performed taking an in-car listening environment as the use case, characterized by multiple loudspeakers, thus, multiple impulse responses (IR). This technique has been compared with a previous heuristic method, achieving superior results.
{"title":"Gravitational Search Algorithm for IIR Filter-Based Audio Equalization","authors":"Giovanni Pepe, L. Gabrielli, S. Squartini, L. Cattani, Carlo Tripodi","doi":"10.23919/Eusipco47968.2020.9287421","DOIUrl":"https://doi.org/10.23919/Eusipco47968.2020.9287421","url":null,"abstract":"In this paper we present an evolutionary algorithm for the design of stable IIR filters for binaural audio equalization. The filters are arranged as a cascade of second-order sections (SOS’s) and the gravitational search algorithm (GSA) is used. This process seeks for optimal coefficients based on a fitness function, possibly leading to unstable filters. To avoid this, we propose two alternative methods. Experiments have been performed taking an in-car listening environment as the use case, characterized by multiple loudspeakers, thus, multiple impulse responses (IR). This technique has been compared with a previous heuristic method, achieving superior results.","PeriodicalId":6705,"journal":{"name":"2020 28th European Signal Processing Conference (EUSIPCO)","volume":"2 1","pages":"496-500"},"PeriodicalIF":0.0,"publicationDate":"2021-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85662086","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-24DOI: 10.23919/Eusipco47968.2020.9287636
Shahab Pasha, J. Lundgren, C. Ritz
This paper reports on the implementation of a multi-channel electronic stethoscope designed to isolate the heart sound from the interfering sounds of the lungs and blood vessels. The multi-channel stethoscope comprises four piezo contact microphones arranged in rectangular and linear arrays. Beamforming and channel equalisation techniques are applied to the multi-channel recordings made in the aortic, pulmonary, tricuspid, and mitral valve areas. The proposed channel equaliser cancels out the distorting effect of the chest and rib cage on the heart sound frequency spectrum. It is shown that the applied beamforming methods effectively suppress the interfering lung noise and improve the signal to interference and noise ratio by 16 dB. The results confirm the superior performance of the implemented multi-channel stethoscope compared with commercially available single-channel electronic stethoscopes.
{"title":"Multi-Channel Electronic Stethoscope for Enhanced Cardiac Auscultation using Beamforming and Equalisation Techniques","authors":"Shahab Pasha, J. Lundgren, C. Ritz","doi":"10.23919/Eusipco47968.2020.9287636","DOIUrl":"https://doi.org/10.23919/Eusipco47968.2020.9287636","url":null,"abstract":"This paper reports on the implementation of a multi-channel electronic stethoscope designed to isolate the heart sound from the interfering sounds of the lungs and blood vessels. The multi-channel stethoscope comprises four piezo contact microphones arranged in rectangular and linear arrays. Beamforming and channel equalisation techniques are applied to the multi-channel recordings made in the aortic, pulmonary, tricuspid, and mitral valve areas. The proposed channel equaliser cancels out the distorting effect of the chest and rib cage on the heart sound frequency spectrum. It is shown that the applied beamforming methods effectively suppress the interfering lung noise and improve the signal to interference and noise ratio by 16 dB. The results confirm the superior performance of the implemented multi-channel stethoscope compared with commercially available single-channel electronic stethoscopes.","PeriodicalId":6705,"journal":{"name":"2020 28th European Signal Processing Conference (EUSIPCO)","volume":"4 1","pages":"1289-1293"},"PeriodicalIF":0.0,"publicationDate":"2021-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85755734","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-24DOI: 10.23919/Eusipco47968.2020.9287810
M. Togami
In this paper, we propose a multi-channel speech source separation technique which connects an unsupervised spatial filtering without a deep neural network (DNN) to a DNN-based speech source separation in a cascade manner. In the speech source separation technique, estimation of a covariance matrix is a highly important part. Recent studies showed that it is effective to estimate the covariance matrix by multiplying cross-correlation of microphone input signal with a time-frequency mask (TFM) inferred by the DNN. However, this assumption is not valid actually and overlapping of multiple speech sources lead to degradation of estimation accuracy of the multi-channel covariance matrix. Instead, we propose a multichannel covariance matrix estimation technique which estimates the covariance matrix by a TFM for the separated speech signal by the unsupervised spatial filtering. Pre-filtered signal can reduce overlapping of multiple speech sources and increase estimation accuracy of the covariance matrix. Experimental results show that the proposed estimation technique of the multichannel covariance matrix is effective.
{"title":"Deep Multi-channel Speech Source Separation with Time-frequency Masking for Spatially Filtered Microphone Input Signal","authors":"M. Togami","doi":"10.23919/Eusipco47968.2020.9287810","DOIUrl":"https://doi.org/10.23919/Eusipco47968.2020.9287810","url":null,"abstract":"In this paper, we propose a multi-channel speech source separation technique which connects an unsupervised spatial filtering without a deep neural network (DNN) to a DNN-based speech source separation in a cascade manner. In the speech source separation technique, estimation of a covariance matrix is a highly important part. Recent studies showed that it is effective to estimate the covariance matrix by multiplying cross-correlation of microphone input signal with a time-frequency mask (TFM) inferred by the DNN. However, this assumption is not valid actually and overlapping of multiple speech sources lead to degradation of estimation accuracy of the multi-channel covariance matrix. Instead, we propose a multichannel covariance matrix estimation technique which estimates the covariance matrix by a TFM for the separated speech signal by the unsupervised spatial filtering. Pre-filtered signal can reduce overlapping of multiple speech sources and increase estimation accuracy of the covariance matrix. Experimental results show that the proposed estimation technique of the multichannel covariance matrix is effective.","PeriodicalId":6705,"journal":{"name":"2020 28th European Signal Processing Conference (EUSIPCO)","volume":"169 1","pages":"266-270"},"PeriodicalIF":0.0,"publicationDate":"2021-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78576008","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-24DOI: 10.23919/Eusipco47968.2020.9287804
D. Luengo, J. Vía, T. Trigano
In this paper, we describe an efficient iterative algorithm for finding sparse solutions to a linear system. Apart from the well-known L1 norm regularization, we introduce an additional cost term promoting solutions without too-close activations. This additional term, which is expressed as a sum of cross-products of absolute values, makes the problem non-convex and difficult to solve. However, the application of the successive convex approximations approach allows us to obtain an efficient algorithm consisting in the solution of a sequence of iteratively reweighted LASSO problems. Numerical simulations on randomly generated waveforms and ECG signals show the good performance of the proposed method.
{"title":"Efficient Iteratively Reweighted LASSO Algorithm for Cross-Products Penalized Sparse Solutions","authors":"D. Luengo, J. Vía, T. Trigano","doi":"10.23919/Eusipco47968.2020.9287804","DOIUrl":"https://doi.org/10.23919/Eusipco47968.2020.9287804","url":null,"abstract":"In this paper, we describe an efficient iterative algorithm for finding sparse solutions to a linear system. Apart from the well-known L1 norm regularization, we introduce an additional cost term promoting solutions without too-close activations. This additional term, which is expressed as a sum of cross-products of absolute values, makes the problem non-convex and difficult to solve. However, the application of the successive convex approximations approach allows us to obtain an efficient algorithm consisting in the solution of a sequence of iteratively reweighted LASSO problems. Numerical simulations on randomly generated waveforms and ECG signals show the good performance of the proposed method.","PeriodicalId":6705,"journal":{"name":"2020 28th European Signal Processing Conference (EUSIPCO)","volume":"15 1","pages":"2045-2049"},"PeriodicalIF":0.0,"publicationDate":"2021-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85428974","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-24DOI: 10.23919/Eusipco47968.2020.9287803
R. Pöhlmann, Siwei Zhang, A. Dammann, P. Hoeher
Both direction-of-arrival (DoA) and direction-of-departure (DoD) of a radio signal contain valuable information for localization. Their estimation with antenna arrays is well known. More recently, multi-mode antennas (MMAs), building on the theory of characteristic modes, have been investigated for DoA estimation. This paper introduces joint DoA and DoD estimation with a single MMA on transmitter and receiver side. In general, the polarization of a signal transmitted by an MMA varies with the direction, which makes an appropriate signal model necessary. For best performance, optimized transmit beamforming should be performed. We derive the Cramér-Rao bound (CRB) for DoA and DoD estimation with MMAs, propose an optimized beamformer (OBF), which minimizes the CRB, and evaluate its performance.
{"title":"Manifold Optimization Based Beamforming for DoA and DoD Estimation with a Single Multi-Mode Antenna","authors":"R. Pöhlmann, Siwei Zhang, A. Dammann, P. Hoeher","doi":"10.23919/Eusipco47968.2020.9287803","DOIUrl":"https://doi.org/10.23919/Eusipco47968.2020.9287803","url":null,"abstract":"Both direction-of-arrival (DoA) and direction-of-departure (DoD) of a radio signal contain valuable information for localization. Their estimation with antenna arrays is well known. More recently, multi-mode antennas (MMAs), building on the theory of characteristic modes, have been investigated for DoA estimation. This paper introduces joint DoA and DoD estimation with a single MMA on transmitter and receiver side. In general, the polarization of a signal transmitted by an MMA varies with the direction, which makes an appropriate signal model necessary. For best performance, optimized transmit beamforming should be performed. We derive the Cramér-Rao bound (CRB) for DoA and DoD estimation with MMAs, propose an optimized beamformer (OBF), which minimizes the CRB, and evaluate its performance.","PeriodicalId":6705,"journal":{"name":"2020 28th European Signal Processing Conference (EUSIPCO)","volume":"72 1","pages":"1841-1845"},"PeriodicalIF":0.0,"publicationDate":"2021-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85846808","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：