复谱映射下融合高效Transformer的语音增强方法

针对卷积神经网络(Convolutional Neural Network,CNN)过去在语音增强中表现优异但对全局特征捕获不足,以及Transformer近年展现出长序列间依赖优势但又存在局部细节特征丢失、参数量大等问题,该文为了充分利用CNN与Transformer的优势并弥补各自不足,提出了一种在复频谱映射下的新型卷积模块与高效Transformer融合的单通道语音增强网络。该网络由编码层、传输层与双分支解码层组成:在编解码部分设计了一种协作学习模块(Collaborative Learning Block,CLB)来监督交互信息,在减少参数量的同时提高主干网络对复特征的获取能力;传输层中则提出一种时频空间注意Transformer模块分别对语音子频带和全频带信息建模,充分利用声学特性来模拟局部频谱模式并捕获谐波间依赖关系。将该模块进一步与通道注意分支相结合,设计了一种可学习的双分支注意融合(Dual-branch Attention Fusion,DAF)机制,从空间-通道角度提取上下文特征以加强信息的多维度传输;最后,在此基础上搭建一种高斯加权渐进网络作为中间传输层,通过堆叠DAF模块进行加权求和后输出以充分利用深层特征,使得解码过程更具鲁棒性。分别在英文VoiceBank-DEMAND数据集、中文THCHS30语料库与115种环境噪声下进行消融以及综合对比实验,结果表明,该文方法仅以最小0.68×10^(6)的参数量,相比于大部分最新相关网络模型取得了更优的主、客观指标,具有较为突出的增强性能与泛化能力。

Spectral matching algorithm based on nonsubsampled contourlet transform and scale-invariant feature transform

A new spectral matching algorithm is proposed by us- ing nonsubsampled contourlet transform and scale-invariant fea- ture transform. The nonsubsampled contourlet transform is used to decompose an image into a low frequency image and several high frequency images, and the scale-invariant feature transform is employed to extract feature points from the low frequency im- age. A proximity matrix is constructed for the feature points of two related images. By singular value decomposition of the proximity matrix, a matching matrix （or matching result） reflecting the match- ing degree among feature points is obtained. Experimental results indicate that the proposed algorithm can reduce time complexity and possess a higher accuracy.

Spectral transformation-based technique for reducing effect of limited pre-correlation bandwidth in the GNSS receiver filter in presence of noise and multipath

This paper proposes an efficient scheme to reduce the pre-correlation bandwidth effect in the global navigation satellite system(GNSS)receiver filtering process.It is mainly based on the application of a spectral transformation to the satellite-emitted signal that effectively reduces its band.At the receiver's end,this operation causes the spreading of noise over a much wider band than that used by the radio frequency stage.Consequently,the resulting auto-correlation function in the acquisition process acquires properties that enhance considerably the performance of the receiver in the presence of the multipath and noise disturbing phenomena.The simulation results demonstrate that the proposed method is a plausible solution for both multipath and noise problems in the GNSS applications for any limited value of the pre-correlation bandwidth in the receiver filter.

Condition Evaluation in Steel Truss Bridge with Fused Hilbert Transform,Spectral Kurtosis,and Bandpass Filter

This study is concerned with the diagnosis of discrepancies in a steel truss bridge by identifying dynamic properties from the vibration response signals of the bridges.The vibration response signals collected at bridges under three different vehicular speeds of 10 km/hr,20 km/hr,and 30 km/hr are analyzed using statistical features such as kurtosis,magnitude of peak-to-peak,root mean square,crest factor as well as impulse factor in time domain,and Stockwell transform in the time-frequency domain.The considered statistical features except for kurtosis show uncertain behavior.The Stockwell transform showed low-resolution outcomes when the presence of noise in the recorded vibration responses.The elimination of noise and extraction of meaningful dynamic properties from the vibration responses is done by applying a new method which comes from the fusion of Hilbert transform with Spectral kurtosis and bandpass filtering.The outcomes obtained from Hilbert transform processed residual signals which are further filtered using bandpass filter show more robustness and accuracy in characterizing bridge modal frequencies from the noisy vibration responses.The proposed method produces a high-resolution frequency response which can unveil the joint discrepancy in the bridge structure.

高光谱图像去噪的稀疏空谱Transformer模型

现阶段Transformer模型的应用提升了高光谱图像去噪的性能,但原始Transformer模型对图像空间-光谱耦合关联性的利用仍存在不足;对空间特征的处理存在过于平滑,容易丢失小尺度结构的现象;同时在光谱维度上也过于关注全部通道特征,缺乏对不同光谱波段间差异性的利用;为了应对这些问题,本文提出了一种新的稀疏空谱Transformer模型,提升了对空谱耦合关联性的利用。在空间维度,引入局部增强模块增强空间特征细节,应对过平滑问题;同时在光谱维度上提出了Top-k稀疏自注意力机制,自适应选择前K个最相关的光谱通道特征进行特征交互,从而能够有效捕获空谱特征。最终通过稀疏空谱Transformer的层级残差连接实现高光谱图像的去噪。在ICVL数据集上分别对高斯噪声和复杂噪声进行去噪处理,峰值信噪比分别达到40.56 dB和40.19 dB,证明了本文提出的稀疏空谱Transformer模型优越的性能。

A SPECTRAL ESTIMATION ALGORITHM USING THE HOUSEHOLDER TRANSFORM

Householder transform is used to triangularize the data matrix, which is basedon the near prediction error equation. It is proved that the sum of squared residuals for eachAR order can be obtained by the main diagonal elements of upper triangular matrix, so thecolumn by column procedure can be used to develop a recursive algorithm for AR modeling andspectral estimation. In most cases, the present algorithm yields the same results as the covariancemethod or modified covariance method does. But in some special cases where the numerical ill-conditioned problems are so serious that the covariance method and modified covariance methodfail to estimate AR spectrum, the presented algorithm still tends to keep good performance. Thetypical computational results are presented finally.

A Short Note on a Differential-Difference Gauge Transformation and a New Spectral Problem

We show that a class of spectral problems are related to the spectral problem of the Volterra lattice through a gauge transformation. The transformation is given. We hope that our discussion can draw attention to the study of gauge transformation theory of differential-difference integrable systems.

A method to compress vibration signals using wavelet packet transformation combined with sub-band vector quantization

A novel compression method for mechanical vibrating signals,binding with sub-band vector quantization(SVQ) by wavelet packet transformation(WPT) and discrete cosine transformation(DCT) is proposed.Firstly,the vibrating signal is decomposed into sub-bands by WPT.Then DCT and adaptive bit allocation are done per sub-band and SVQ is performed in each sub-band.It is noted that,after DCT,we only need to code the first components whose numbers are determined by the bits allocated to that sub-band.Through an actual signal,our algorithm is proven to improve the signal-to-noise ratio(SNR) of the reconstructed signal effectively,especially in the situation of lowrate transmission.

Spectrally Efficient Multi-Carrier Modulation Using Gabor Transform

Non Orthogonal Frequency Division Multiplexing (NOFDM) systems make use of a transmission signal set which is not restricted to orthonormal bases unlike previous OFDM systems. The usage of non-orthogonal bases generally results in a trade-off between Bit Error Rate (BER) and receiver complexity. This paper studies the use of Gabor based on designing a Spectrally Efficient Multi-Carrier Modulation Scheme. Using Gabor Transform with a specific Gaussian envelope;we derive the expected BER-SNR performance. The spectral usage of such a NOFDM system when affected by a channel that imparts Additive White Gaussian Noise (AWGN) is estimated. We compare the obtained results with an OFDM system and observe that with comparable BER performance, this system gives a better spectral usage. The effect of window length on spectral usage is also analyzed.

Efficient simulation of spatially correlated non-stationary ground motions by wavelet-packet algorithm and spectral representation method

Although the classical spectral representation method(SRM)has been widely used in the generation of spatially varying ground motions,there are still challenges in efficient simulation of the non-stationary stochastic vector process in practice.The first problem is the inherent limitation and inflexibility of the deterministic time/frequency modulation function.Another difficulty is the estimation of evolutionary power spectral density(EPSD)with quite a few samples.To tackle these problems,the wavelet packet transform(WPT)algorithm is utilized to build a time-varying spectrum of seed recording which describes the energy distribution in the time-frequency domain.The time-varying spectrum is proven to preserve the time and frequency marginal property as theoretical EPSD will do for the stationary process.For the simulation of spatially varying ground motions,the auto-EPSD for all locations is directly estimated using the time-varying spectrum of seed recording rather than matching predefined EPSD models.Then the constructed spectral matrix is incorporated in SRM to simulate spatially varying non-stationary ground motions using efficient Cholesky decomposition techniques.In addition to a good match with the target coherency model,two numerical examples indicate that the generated time histories retain the physical properties of the prescribed seed recording,including waveform,temporal/spectral non-stationarity,normalized energy buildup,and significant duration.

Performance of Continuous Wavelet Transform over Fourier Transform in Features Resolutions

This study presents a comparative analysis of two image enhancement techniques, Continuous Wavelet Transform (CWT) and Fast Fourier Transform (FFT), in the context of improving the clarity of high-quality 3D seismic data obtained from the Tano Basin in West Africa, Ghana. The research focuses on a comparative analysis of image clarity in seismic attribute analysis to facilitate the identification of reservoir features within the subsurface structures. The findings of the study indicate that CWT has a significant advantage over FFT in terms of image quality and identifying subsurface structures. The results demonstrate the superior performance of CWT in providing a better representation, making it more effective for seismic attribute analysis. The study highlights the importance of choosing the appropriate image enhancement technique based on the specific application needs and the broader context of the study. While CWT provides high-quality images and superior performance in identifying subsurface structures, the selection between these methods should be made judiciously, taking into account the objectives of the study and the characteristics of the signals being analyzed. The research provides valuable insights into the decision-making process for selecting image enhancement techniques in seismic data analysis, helping researchers and practitioners make informed choices that cater to the unique requirements of their studies. Ultimately, this study contributes to the advancement of the field of subsurface imaging and geological feature identification.

基于频谱特征混合Transformer的红外和可见光图像融合

为了解决传统红外与可见光图像融合方法对细节与频率信息表征能力不足、融合结果存在模糊伪影的问题,提出一种基于频谱特征混合Transformer的红外和可见光图像融合算法。在Transformer的基础上,利用傅里叶变换将图像域特征映射到频域,设计了一种新的复数Transformer来提取源图像的深层频域信息,并与图像域特征进行混合,以此提高网络对细节与频率信息的表征能力。此外,在图像重建前设计了一种新的令牌替换模块,动态评估Transformer令牌的显著性后消除得分较低的令牌,防止融合图像出现伪影。在MSRS数据集上进行的定性和定量实验结果显示,与九种先进的算法相比,该算法具有较好的融合效果。

The optimal fractional Gabor transform based on the adaptive window function and its application

We designed the window function of the optimal Gabor transform based on the time-frequency rotation property of the fractional Fourier transform. Thus, we obtained the adaptive optimal Gabor transform in the fractional domain and improved the time-frequency concentration of the Gabor transform. The algorithm first searches for the optimal rotation factor, then performs the p-th FrFT of the signal and, finally, performs time and frequency analysis of the FrFT result. Finally, the algorithm rotates the plane in the fractional domain back to the normal time-frequency plane. This promotes the application of FrFT in the field of high-resolution reservoir prediction. Additionally, we proposed an adaptive search method for the optimal rotation factor using the Parseval principle in the fractional domain, which simplifies the algorithm. We carried out spectrum decomposition of the seismic signal, which showed that the instantaneous frequency slices obtained by the proposed algorithm are superior to the ones obtained by the traditional Gabor transform. The adaptive time frequency analysis is of great significance to seismic signal processing.

Ground roll attenuation using a time-frequency dependent polarization filter based on the S transform

The ground roll and body wave usually show significant differences in arrival time, frequency content, and polarization characteristics, and conventional polarization filters that operate in either the time or frequency domain cannot consider all these elements. Therefore, we have developed a time-frequency dependent polarization filter based on the S transform to attenuate the ground roll in seismic records. Our approach adopts the complex coefficients of the S transform of the multi-component seismic data to estimate the local polarization attributes and utilizes the estimated attributes to construct the filter function. In this study, we select the S transform to design this polarization filter because its scalable window length can ensure the same number of cycles of a Fourier sinusoid, thereby rendering more precise estimation of local polarization attributes. The results of applying our approach in synthetic and real data examples demonstrate that the proposed polarization filter can effectively attenuate the ground roll and successfully preserve the body wave.

Generation of artificial earthquakes for matching target response unsmooth spectrum via wavelet package transform

The seismic records of target response spectrum used in the time-history analysis should be allowed to meet the norms. However, the previous fitting methods of target spectrum are mostly for the situations that the target spectrum is a smooth curve. In many cases, it needs to match unsmooth target spectrum for single determined response spectrum. An adjustment of time history via wavelet packet transform was presented, which is able to fit unsmooth target spectrum. It was found that there is a certain bias between the band center frequency of the component of seismic record after wavelet packet decomposition and the peak frequency of response spectra of wavelet packet components. For this reason, five strategies were presented to select iteration points, and the effects of the five strategies were compared with two calculation examples. It was turned out that the peak frequency of the response spectrum of wavelet packet component can lead to good fitting effect when it is selected as the iteration point. In the iteration process, it shows great promise in fitting non-smooth target spectrum and has a trend of converge.

面向高光谱医学图像分类的空-谱自注意力Transformer

高光谱成像技术的飞速发展给非侵入式医学成像带来新的契机,但高光谱医学图像具有高维度、高冗余以及“图谱合一”的特点,亟需针对上述特点设计智能诊断算法。近年来,Transformer已经在高光谱医学图像处理领域得到广泛应用。然而,不同仪器设备、不同采集操作所获得的高光谱医学图像差异较大,这给现有Transformer诊断模型的实际应用带来了巨大挑战。针对上述问题,本文提出了一种空-谱自注意力Transformer(S3AT),自适应挖掘像素与像素间、波段与波段间的内蕴联系,并在分类阶段融合多个视野下的预测结果。首先,在Transformer编码器中,设计一种空-谱自注意力机制,获取不同视野下高光谱图像上的关键空间信息和重要波段,并将不同视野下所获得的空-谱自注意力进行融合。其次,在模型分类阶段,将不同视野下的预测结果根据可学习权重进行加权融合,对图像进行综合预测。在In-vivo Human Brain和BloodCell HSI两个数据集上,本文算法总体分类精度分别达到82.25%和91.74%。实验结果表明,所提出的算法有效改善高光谱医学图像分类性能。

Q-factor estimation in CMP gather and the continuous spectral ratio slope method

The attenuation factor or quality factor(Q-factor or Q) has been used to measure the energy attenuation of seismic waves propagating in underground media. Many methods are used to estimate the Q-factor. We propose a method to calculate the Q-factor based on the prestack Q-factor inversion and the generalized S-transform. The proposed method specifies a standard primary wavelet and calculates the cumulative Q-factors; then, it finds the interlaminar Q-factors using the relation between Q and offset(QVO) and the Dix formula. The proposed method is alternative to methods that calculate interlaminar Q-factors after horizon picking. Because the frequency spectrum of each horizon can be extracted continuously on a 2D time–frequency spectrum, the method is called the continuous spectral ratio slope(CSRS) method. Compared with the other Q-inversion methods, the method offers nearly effortless computations and stability, and has mathematical and physical significance. We use numerical modeling to verify the feasibility of the method and apply it to real data from an oilfield in Ahdeb, Iraq. The results suggest that the resolution and spatial stability of the Q-profile are optimal and contain abundant interlaminar information that is extremely helpful in making lithology and fluid predictions.

Application of Wavelet Algorithm to Spectral Analysis of Oceanic Waves and Offshore Structure Responses

Fourier transform （FF） is a commonly used method in spectral analysis of ocean wave and offshore structure responses, but it is not suitable for records of short length. In this paper another method, wavelet transform （WT）, is applied to ＇analyze the data of short length. The Morlet wavelet is employed to calculate the spectra density functions for wave records and simulated Floating Production Storage and Offloading （FPSO） vessels＇ responses. Computed wave data include simulated wave data based on JONSWAP spectrum and the recorded data of Storm 149 from North Alwyn. Wavelet method is validated by comparing the statistical characteristics by WF method and those by fast Fourier transform （FFT） method with those of target spectra. The spectral density fnnctions＇ shapes calculated by WT are less malformed and have less error of statistical characteristics compared with those by FT especially when the record lengths decrease.

Constant envelope FrFT OFDM: spectral and energy efficiency analysis

Constant envelope with a fractional Fourier transformorthogonal frequency division multiplexing(CE-FrFT-OFDM)is a special case of a constant envelope OFDM(CE-OFDM),both being energy efficient wireless communication techniques with a 0 dB peak to average power ratio(PAPR).However,with the proper selection of fractional order,the first technique has a high bit error rate(BER)performance in the frequency-time selective channels.This paper performs further analysis of CE-FrFT-OFDM by examining its spectral efficiency(SE)and energy efficiency(EE)and compare to the famous OFDM and FrFT-OFDM techniques.Analytical and comprehensive simulations conducted show that,the CE-FrFT-OFDM has five times the EE of OFDM and FrFT-OFDM systems with a slightly less SE.Increasing CE-FrFT-OFDM’s transmission power by increasing its amplitude to 1.7 increases its SE to match that of the OFDM and FrFT-OFDM systems while slightly reducing its EE by 20%to be four times that of OFDM and FrFTOFDM systems.OFDM and FrFT-OFDM’s amplitude fluctuations cause rapid changing output back-off(OBO)power requirements and further reduce power amplifier(PA)efficiency while CE-FrFTOFDM stable operational linear range makes it a better candidate and outperforms the other techniques when their OBO exceeds 1.7.Higher EE and low BER in time-frequency selective channel are attracting features for CE-FrFT-OFDM deployment in mobile devices.

Denoising of seismic data via multi-scale ridgelet transform

Noise has traditionally been suppressed or eliminated in seismic data sets by the use of Fourier filters and, to a lesser degree, nonlinear statistical filters. Although these methods are quite useful under specific conditions, they may produce undesirable effects for the low signal to noise ratio data. In this paper, a new method, multi-scale ridgelet transform, is used in the light of the theory of ridgelet transform. We employ wavelet transform to do sub-band decomposition for the signals and then use non-linear thresholding in ridgelet domain for every block. In other words, it is based on the idea of partition, at sufficiently fine scale, a curving singularity looks straight, and so ridgelet transform can work well in such cases. Applications on both synthetic data and actual seismic data from Sichuan basin, South China, show that the new method eliminates the noise portion of the signal more efficiently and retains a greater amount of geologic data than other methods, the quality and consecutiveness of seismic event are improved obviously as well as the quality of section is improved.