Aiming at the poor performance of speech signal detection at low signal-to-noise ratio(SNR),a method is proposed to detect active speech frames based on multi-window time-frequency(T-F)diagrams.First,the T-F diagram o...Aiming at the poor performance of speech signal detection at low signal-to-noise ratio(SNR),a method is proposed to detect active speech frames based on multi-window time-frequency(T-F)diagrams.First,the T-F diagram of the signal is calculated based on a multi-window T-F analysis,and a speech test statistic is constructed based on the characteristic difference between the signal and background noise.Second,the dynamic double-threshold processing is used for preliminary detection,and then the global double-threshold value is obtained using K-means clustering.Finally,the detection results are obtained by sequential decision.The experimental results show that the overall performance of the method is better than that of traditional methods under various SNR conditions and background noises.This method also has the advantages of low complexity,strong robustness,and adaptability to multi-national languages.展开更多
For many water quality studies,a data analyst,or modeler, may need to know the spatio-temproal patterns of the data sets and their relationships in both pre-processing and post-processing. Geographic Information Syste...For many water quality studies,a data analyst,or modeler, may need to know the spatio-temproal patterns of the data sets and their relationships in both pre-processing and post-processing. Geographic Information System(GIS) can provide an exploratory spat展开更多
为了提升运动想象脑电(MI-EEG)信号的分类精度,提出多尺度滑窗注意力时序卷积网络(MSWATCN),充分挖掘MI-EEG信号的时空信息.结合多尺度双流分组卷积、滑动窗口多头注意力机制和窗口化时间卷积模块,实现对MI-EEG信号复杂时空特性的精准解...为了提升运动想象脑电(MI-EEG)信号的分类精度,提出多尺度滑窗注意力时序卷积网络(MSWATCN),充分挖掘MI-EEG信号的时空信息.结合多尺度双流分组卷积、滑动窗口多头注意力机制和窗口化时间卷积模块,实现对MI-EEG信号复杂时空特性的精准解码.利用多尺度卷积模块提取信号的底层时空特征,通过滑动窗口注意力机制聚焦局部关键特征,突出对分类任务重要的信息.窗口化时间卷积模块通过建模时间序列中的长期依赖关系,增强模型处理时序信息的能力.实验结果表明,MSWATCN在BCI Competition IV 2a和2b数据集上的分类准确率和一致性优于对比网络和基准模型.展开更多
Given L, N, M ∈ N and an NZ-periodic set S in Z, let l2(S) be the closed subspace of l2(Z) consisting of sequences vanishing outside S. For f = { fl : 0≤l≤L-1 }l2(Z), we denote by G(f, N, M) the Gabor system genera...Given L, N, M ∈ N and an NZ-periodic set S in Z, let l2(S) be the closed subspace of l2(Z) consisting of sequences vanishing outside S. For f = { fl : 0≤l≤L-1 }l2(Z), we denote by G(f, N, M) the Gabor system generated by f, and by L(f, N, M) the closed linear subspace generated by G(f, N, M). This paper addresses density results, frame conditions for a Gabor system G(g, N, M) in l2(S), and Gabor duals of the form G(a, N, M) in some L(h, N, M) for a frame G(g, N, M) in l2(S) (so-called oblique duals). We obtain a density theorem for a Gabor system G(g, N, M) in l2(S), and show that such condition is suficient for theexistence of g={XE1:0≤l≤L-1} with G(g,N,m) We characterize g with G(g,N,m) being respectively a frame for L(g,N,m) being a tight frame for l2(S).and G(h, N, M ) in L(h, N, M ), we establish a criterion for the existence of an oblique Gabor dual of g in L(h, N, M), study the uniqueness of oblique Gabor dual, and derive an explicit expression of a class of oblique Gabor duals (among which the one with the smallest norm is obtained). Some examples are also provided.展开更多
Wavelet and Gabor systems are based on translation-and-dilation and translation-and-modulation operators,respectively,and have been studied extensively.However,dilation-and-modulation systems cannot be derived from wa...Wavelet and Gabor systems are based on translation-and-dilation and translation-and-modulation operators,respectively,and have been studied extensively.However,dilation-and-modulation systems cannot be derived from wavelet or Gabor systems.This study aims to investigate a class of dilation-and-modulation systems in the causal signal space L^2(R+).L^2(R+)can be identified as a subspace of L^2(R),which consists of all L^2(R)-functions supported on R+but not closed under the Fourier transform.Therefore,the Fourier transform method does not work in L^2(R+).Herein,we introduce the notion ofΘa-transform in L^2(R+)and characterize the dilation-and-modulation frames and dual frames in L^2(R+)using theΘa-transform;and present an explicit expression of all duals with the same structure for a general dilation-and-modulation frame for L^2(R+).Furthermore,it has been proven that an arbitrary frame of this form is always nonredundant whenever the number of the generators is 1 and is always redundant whenever the number is greater than 1.Finally,some examples are provided to illustrate the generality of our results.展开更多
针对视频异常事件的时空相关性学习对检测性能存在重要影响的问题,提出了基于融合双支特征的带约束损失的视频异常检测方法(Dual-branch Feature Fusion Based Constrained Multi-loss Video Anomaly Detection,DBF-CML-transMIL)。该...针对视频异常事件的时空相关性学习对检测性能存在重要影响的问题,提出了基于融合双支特征的带约束损失的视频异常检测方法(Dual-branch Feature Fusion Based Constrained Multi-loss Video Anomaly Detection,DBF-CML-transMIL)。该方法考虑多示例学习中片段的显著性和相关性,利用多层线性神经网络学习各片段的空间显著性特征,并设计级联Transformer融合模块来学习示例间的多层时序相关性;然后利用多损失模型对融合特征进行多loss监督学习,以丰富预测的多样性;针对现有top-k的离散性问题,提出了带约束机制的滑窗top-k强化异常事件的相关性。在ShanghaiTech和UCF-Crime数据集上的对比实验与消融实验表明,DBF-CML-transMIL的异常检测曲线下面积(Area Under Curve,AUC)分别达到97.33%和83.82%;各模块都能有效提升视频异常事件检测的性能。展开更多
基金The National Natural Science Foundation of China(No.12174053,91938203,11674057,11874109)the Fundamental Research Funds for the Central Universities(No.2242021k30019).
文摘Aiming at the poor performance of speech signal detection at low signal-to-noise ratio(SNR),a method is proposed to detect active speech frames based on multi-window time-frequency(T-F)diagrams.First,the T-F diagram of the signal is calculated based on a multi-window T-F analysis,and a speech test statistic is constructed based on the characteristic difference between the signal and background noise.Second,the dynamic double-threshold processing is used for preliminary detection,and then the global double-threshold value is obtained using K-means clustering.Finally,the detection results are obtained by sequential decision.The experimental results show that the overall performance of the method is better than that of traditional methods under various SNR conditions and background noises.This method also has the advantages of low complexity,strong robustness,and adaptability to multi-national languages.
文摘For many water quality studies,a data analyst,or modeler, may need to know the spatio-temproal patterns of the data sets and their relationships in both pre-processing and post-processing. Geographic Information System(GIS) can provide an exploratory spat
文摘为了提升运动想象脑电(MI-EEG)信号的分类精度,提出多尺度滑窗注意力时序卷积网络(MSWATCN),充分挖掘MI-EEG信号的时空信息.结合多尺度双流分组卷积、滑动窗口多头注意力机制和窗口化时间卷积模块,实现对MI-EEG信号复杂时空特性的精准解码.利用多尺度卷积模块提取信号的底层时空特征,通过滑动窗口注意力机制聚焦局部关键特征,突出对分类任务重要的信息.窗口化时间卷积模块通过建模时间序列中的长期依赖关系,增强模型处理时序信息的能力.实验结果表明,MSWATCN在BCI Competition IV 2a和2b数据集上的分类准确率和一致性优于对比网络和基准模型.
基金supported by National Natural Science Foundation of China (Grant Nos. 10901013, 10671008)Beijing Natural Science Foundation (Grant No. 1092001)+1 种基金the Scientific Research Common Program of Beijing Municipal Commission of Education (Grant No. KM201110005030)the Project Sponsored by SRF for ROCS, SEM of China
文摘Given L, N, M ∈ N and an NZ-periodic set S in Z, let l2(S) be the closed subspace of l2(Z) consisting of sequences vanishing outside S. For f = { fl : 0≤l≤L-1 }l2(Z), we denote by G(f, N, M) the Gabor system generated by f, and by L(f, N, M) the closed linear subspace generated by G(f, N, M). This paper addresses density results, frame conditions for a Gabor system G(g, N, M) in l2(S), and Gabor duals of the form G(a, N, M) in some L(h, N, M) for a frame G(g, N, M) in l2(S) (so-called oblique duals). We obtain a density theorem for a Gabor system G(g, N, M) in l2(S), and show that such condition is suficient for theexistence of g={XE1:0≤l≤L-1} with G(g,N,m) We characterize g with G(g,N,m) being respectively a frame for L(g,N,m) being a tight frame for l2(S).and G(h, N, M ) in L(h, N, M ), we establish a criterion for the existence of an oblique Gabor dual of g in L(h, N, M), study the uniqueness of oblique Gabor dual, and derive an explicit expression of a class of oblique Gabor duals (among which the one with the smallest norm is obtained). Some examples are also provided.
基金supported by National Natural Science Foundation of China(Grant No.11271037)。
文摘Wavelet and Gabor systems are based on translation-and-dilation and translation-and-modulation operators,respectively,and have been studied extensively.However,dilation-and-modulation systems cannot be derived from wavelet or Gabor systems.This study aims to investigate a class of dilation-and-modulation systems in the causal signal space L^2(R+).L^2(R+)can be identified as a subspace of L^2(R),which consists of all L^2(R)-functions supported on R+but not closed under the Fourier transform.Therefore,the Fourier transform method does not work in L^2(R+).Herein,we introduce the notion ofΘa-transform in L^2(R+)and characterize the dilation-and-modulation frames and dual frames in L^2(R+)using theΘa-transform;and present an explicit expression of all duals with the same structure for a general dilation-and-modulation frame for L^2(R+).Furthermore,it has been proven that an arbitrary frame of this form is always nonredundant whenever the number of the generators is 1 and is always redundant whenever the number is greater than 1.Finally,some examples are provided to illustrate the generality of our results.
文摘针对视频异常事件的时空相关性学习对检测性能存在重要影响的问题,提出了基于融合双支特征的带约束损失的视频异常检测方法(Dual-branch Feature Fusion Based Constrained Multi-loss Video Anomaly Detection,DBF-CML-transMIL)。该方法考虑多示例学习中片段的显著性和相关性,利用多层线性神经网络学习各片段的空间显著性特征,并设计级联Transformer融合模块来学习示例间的多层时序相关性;然后利用多损失模型对融合特征进行多loss监督学习,以丰富预测的多样性;针对现有top-k的离散性问题,提出了带约束机制的滑窗top-k强化异常事件的相关性。在ShanghaiTech和UCF-Crime数据集上的对比实验与消融实验表明,DBF-CML-transMIL的异常检测曲线下面积(Area Under Curve,AUC)分别达到97.33%和83.82%;各模块都能有效提升视频异常事件检测的性能。
