Short-time Fourier Transform Research Articles

IMAR in Frequency Domain for Separating Speech Signals by STFT

Blind Source Separation is a statistical method for isolating signals coming from unidentified sources that have been picked up by a variety of sensors. In the fields of biological signal analysis and speech analysis technology, it is an active researcher. The quantity of noise that is present in the signal throughout the speech signal separation process is a disadvantage. This noise affects the signal that has been separated, and it is often loud melodic noise. For the purpose of separating speech signals from BSS, the suggested method makes use of a network that is configured with the parameters of Instantaneous Mixing in addition to an Auto Regressive model and a maximum-likelihood function. In order to separate the individual voice signal from the input mixed signals, the Back Propagation Network is put into operation. The Short Time Fourier Transform is then utilized for the purpose of dividing the spoken stream into extremely precise time periods. Next, Maximum-likelihood method is implemented to determine the optimum values of the IMAR model's W and G parameters. This step is done after the ideal values have been estimated. The suggested model was tested using a combination of voice signals and microphone signals, and the results demonstrated that it provided superior execution in comparison to other algorithms that are currently in use.

Research on the characteristics of EV interior sound quality and its dynamic active control system design and development un-der accelerated driving conditions.

In order to improve the interior sound quality of Electric Vehicles (EV), solve the problem of low sense of power and comfort of the interior sound as well as the large electromagnetic excitation order noise of motor and the sharp interior sound, this article designs a dynamic active sound control system for EV under accelerated driving conditions. Firstly, by comparing and analyzing the sound spectrum characteristics of fuel vehicle (FV) and EV during acceleration, a short-time Fourier transform (STFT) is adopted to extract and synthesize the engine sound. Secondly, the influence of the engine order composition and the energy distribution in the frequency domain on the sound quality of the vehicle is analyzed, and an active control system for sound quality is proposed. And the software and hardware development of the active control sound system is completed. Finally, through real-vehicle testing and verification, the sense of comfort and power of the EV interior sound has been greatly improved during acceleration, and the total value of interior sound can meet the requirement. The sound pressure level and loudness of interior sound have been increased, and the sharpness of the sound inside the vehicle has been improved, with a maximum reduction of 1.0acum.

Performance Analysis of Feature Mel Frequency Cepstral Coefficient and Short Time Fourier Transform Input for Lie Detection using Convolutional Neural Network

This study aims to determine which model is more effective in detecting lies between models with Mel Frequency Cepstral Coefficient (MFCC) and Short Time Fourier Transform (STFT) processes using Convolutional Neural Network (CNN). MFCC and STFT processes are based on digital voice data from video recordings that have been given lie or truth information regarding certain situations. Data is then pre-processed and trained on CNN. The results of model performance evaluation with hyper-tuning parameters and random search implementation show that using MFCC as Voice data processing provides better performance with higher accuracy than using the STFT process. The best parameters from MFCC are obtained with filter convolutional=64, kerneconvolutional1=5, filterconvolutional2=112, kernel convolutional2=3, filter convolutional3=32, kernelconvolutional3 =5, dense1=96, optimizer=RMSProp, learning rate=0.001 which achieves an accuracy of 97.13%, with an AUC value of 0.97. Using the STFT, the best parameters are obtained with filter convolutional1=96, kernel convolutional1=5, convolutional2 filters=48, convolutional2 kernels=5, convolutional3 filters=96, convolutional3 kernels=5, dense1=128, Optimizer=Adaddelta, learning rate=0.001, which achieves an accuracy of 95.39% with an AUC value of 0.95. Prosodics are used to compare the performance of MFCC and STFT. The result is that prosodic has a low accuracy of 68%. The analysis shows that using MFCC as the process of sound extraction with the CNN model produces the best performance for cases of lie detection using audio. It can be optimized for further research by combining CNN architectural models such as ResNet, AlexNet, and other architectures to obtain new models and improve lie detection accuracy.

Building structure-borne noise measurements and estimation due to train operations in tunnel

The perception of structure-borne noise is particularly salient when train passes through the tunnel under the buildings, which has a negative impact on human health. In the process of constructing buildings along metro lines, it is crucial to estimate indoor structure-borne noise levels in order to enhance design and prevent any negative impact on human comfort. This study conducted measurements of structure-borne noise, reverberation time, and train-induced vibrations in Guangzhou, China to investigate the generation, propagation, and dissipation mechanisms of structure-borne noise. An approach based on Short-Time Fourier Transform and Schroeder integral was proposed for obtaining frequency-dependent reverberation time. Additionally, a deep learning-based approach incorporating indoor vibrations, frequency-dependent reverberation time, and room parameters as inputs was proposed based on Genetic Algorithm-Artificial Neural Network. The estimated structure-borne noise levels demonstrated good agreement with measured values, indicating the feasibility of the approach. The finding of this research facilitates a clear comprehension of the generation, distribution, and dissipation mechanisms of indoor structure-borne noise for engineers while also enabling convenient acquisition of indoor structure-borne noise. The estimated noise levels can be effectively utilized during building design processes along metro lines to mitigate adverse impacts on human comfort.

The Synchrosqueezed Method and Its Theory-Analysis-Based Novel Short-Time Fractional Fourier Transform for Chirp Signals

Time–frequency analysis is an important tool used for the processing and interpretation of non-stationary signals, such as seismic data and remote sensing data. In this paper, based on the novel short-time fractional Fourier transform (STFRFT), a new modified STFRFT is first proposed which can also generalize the properties of the modified short-time Fourier transform (STFT). Then, in the modified STFRFT domain, we derive the instantaneous frequency estimator for the chirp signal and present a new type of synchrosqueezing STFRFT (FRSST). The proposed FRSST presents many results similar to those of the synchrosqueezing STFT (FSST), and it extends the harmonic signal to a chirp signal that offers attractive new features. Furthermore, we provide a detailed analysis of the signal reconstruction, theories, and some properties of the proposed FRSST. Several experiments are conducted, and all of the results illustrate that the proposed FRSST is more effective than the FSST. Finally, based on the linear amplitude modulation and frequency modulation signal, we present a derivation for analyzing the limitations of the FRSST.

Learning Temporal Resolution in Spectrogram for Audio Classification

The audio spectrogram is a time-frequency representation that has been widely used for audio classification. One of the key attributes of the audio spectrogram is the temporal resolution, which depends on the hop size used in the Short-Time Fourier Transform (STFT). Previous works generally assume the hop size should be a constant value (e.g., 10 ms). However, a fixed temporal resolution is not always optimal for different types of sound. The temporal resolution affects not only classification accuracy but also computational cost. This paper proposes a novel method, DiffRes, that enables differentiable temporal resolution modeling for audio classification. Given a spectrogram calculated with a fixed hop size, DiffRes merges non-essential time frames while preserving important frames. DiffRes acts as a "drop-in" module between an audio spectrogram and a classifier and can be jointly optimized with the classification task. We evaluate DiffRes on five audio classification tasks, using mel-spectrograms as the acoustic features, followed by off-the-shelf classifier backbones. Compared with previous methods using the fixed temporal resolution, the DiffRes-based method can achieve the equivalent or better classification accuracy with at least 25% computational cost reduction. We further show that DiffRes can improve classification accuracy by increasing the temporal resolution of input acoustic features, without adding to the computational cost.

Advancing Cough Classification: Swin Transformer vs. 2D CNN with STFT and Augmentation Techniques

Coughing, a common symptom associated with various respiratory problems, is a crucial indicator for diagnosing and tracking respiratory diseases. Accurate identification and categorization of cough sounds, specially distinguishing between wet and dry coughs, are essential for understanding underlying health conditions. This research focuses on applying the Swin Transformer for classifying wet and dry coughs using short-time Fourier transform (STFT) representations. We conduct a comprehensive evaluation, including a performance comparison with a 2D convolutional neural network (2D CNN) model, and exploration of two distinct image augmentation methods: time mask augmentation and classical image augmentation techniques. Extensive hyperparameter tuning is performed to optimize the Swin Transformer’s performance, considering input size, patch size, embedding size, number of epochs, optimizer type, and regularization technique. Our results demonstrate the Swin Transformer’s superior accuracy, particularly when trained on classically augmented STFT images with optimized settings (320 × 320 input size, RMS optimizer, 8 × 8 patch size, and an embedding size of 128). The approach achieves remarkable testing accuracy (88.37%) and ROC AUC values (94.88%) on the challenging crowdsourced COUGHVID dataset, marking improvements of approximately 2.5% and 11% increases in testing accuracy and ROC AUC values, respectively, compared to previous studies. These findings underscore the efficacy of Swin Transformer architectures in disease detection and healthcare classification problems.

Identification and Analysis of Micro-Doppler Signature of a Bird Versus Micro-UAV

The groundbreaking advancement of micro unmanned aerial vehicles (micro-UAVs) has been staggering. The diversity of micro-UAV operations is demanding in most sectors. However, the current regulatory framework for the civilian use of these devices is still insufficient. The operation of micro-UAVs may pose risks, including privacy violations and collision hazards. To address these concerns, a radar with advanced processing is needed. This study presents a preliminary design of an S-band continuous wave (CW) radar, which was simulated using MATLAB. The size of the rotating propeller blades of the micro-UAV ranges from 20 to 40 cm in length, while the size of a bird’s flapping wing measures 35 cm in length, comprising 22 cm for the upper arm and 13 cm for the lower arm. The analysis was conducted under hovering conditions, where the target's main body is stationary while its micro-parts move continuously. The Short-Time Fourier Transform (STFT) analysis successfully identified the unique signature of both targets. The results showed that the S-band CW radar design at 5 GHz is effective in extracting the micro-Doppler signature of a bird versus a micro-UAV. The extracted features can be used as additional characteristics for target classification in the future.

Analysis of Vocal Signatures of COVID-19 in Cough Sounds: A Newer Diagnostic Approach Using Artificial Intelligence.

Artificial intelligence (AI) based models are explored increasingly in the medical field. The highly contagious pandemic of coronavirus disease 2019 (COVID-19) affected the world and availability of diagnostic tools high resolution computed tomography (HRCT) and/or real-time reverse transcriptase-polymerase chain reaction (RTPCR) was very limited, costly and time consuming. Therefore, the use of AI in COVID-19 for diagnosis using cough sounds can be efficacious and cost effective for screening in clinic or hospital and help in early diagnosis and further management of patients. To develop an accurate and fast voice-processing AI software to determine voice-based signatures in discriminating COVID-19 and non-COVID-19 cough sounds for screening of COVID-19. A prospective study involving 117 patients was performed based on online and/or offline voice data collection of cough sounds of COVID-19 patients in isolation ward of a tertiary care teaching hospital and non-COVID-19 participants using a smart phone. A website-based AI software was developed to identify the cough sounds as COVID-19 or non-COVID-19. The data were divided into three segments including training set, validation set and test set. A pre-processing algorithmwas utilized and combined with Short Time Fourier Transform feature representation and Logistic regression model. A precise software was used to identify vocal signatures and K-fold cross validation was carried out. A total of 117 audio recordings of cough sounds were collected through the developed website after inclusion-exclusion criteria out of which 52 have been marked belonging to COVID-19 positive, while 65 were marked as COVID-19 negative/unsure /never had COVID-19, which were assumed to be COVID-19 negative based on RT-PCR test results. The mean and standard error values for the accuracies attained at the end of each experiment in training, validation and testing set were found to be 67.34%±0.22, 58.57%±1.11 and 64.60%±1.79 respectively. The weight values were found to be positive which were contributing towards predicting the samples as COVID-19 positive with large spikes around 7.5 kHz, 7.8 kHz, 8.6 kHz and 11 kHz which can be used for classification. The proposed AI based approach can be a helpful screening tool for COVID-19 using vocal sounds of cough. It can help the health system by reducing the cost burden and improving overall diagnosis and management of the disease.

Detection of epileptic seizure using EEG signals analysis based on deep learning techniques

The brain neurons' electrical activities represented by Electroencephalogram (EEG) signals are the most common data for diagnosing Epilepsy seizure, which is considered a chronic nervous disorder that cannot be controlled medically using surgical operation or medications with more than 40 % of Epilepsy seizure case. With the progress and development of artificial intelligence and deep learning techniques, it becomes possible to detect these seizures over the observation of the non-stationary-dynamic EEG signals, which contain important information about the mental state of patients. This paper provides a concerted deep machine learning model consisting of two simultaneous techniques detecting the activity of epileptic seizures using EEG signals. The time-frequency image of EEG waves and EEG raw waves are used as input components for the convolution neural network (CNN) and recurrent neural network (RNN) with long- and short-term memory (LSTM). Two processing signal methods have been used, Short-Time Fourier Transform (STFT) and Continuous Wavelet Transformation (CWT), have been used for generating spectrogram and scalogram images with sizes of 77 × 75 and 32 × 32, respectively. The experimental results showed a detection accuracy of 99.57 %, 99.57 % using CWT Scalograms, and 99.26 %, 97.12 % using STFT spectrograms as CNN input for the Bonn University dataset and the CHB-MIT dataset, respectively. Thus, the proposed models provide the ability to detect epileptic seizures with high success compared to previous studies.

Fast adaptive time-varying window length STFT for compound track short-wave defects identification

Track short-wave defect, as a common defect of railway tracks, can cause dynamic response on both vehicle and its components and thus severely affect the stability and safety of running vehicle. The vibration signal, collected by an accelerator mounted on the axle box of vehicle, is a kind of vehicle dynamic response signal. It consists of complex periodic and impact signal components, which poses challenges for track short-wave defects identification. In this paper, the forms and features of vibration signal of vehicle dynamic response are analyzed at first. Then, based on short-time Fourier transform (STFT), a fast adaptive time-varying window length STFT (FATW-STFT) is developed for compound track short-wave defect analysis. Specifically, by quickly calculating a series of time-frequency planes with various window length, modified kurtosis, boundary function, and Rényi entropy are introduced to sift the time-frequency planes and optimize the window lengths at each time point. Unlike other STFT-based methods, FATW-STFT uses fully adaptive window length optimization to distinguish both periodic and impactive signal components, which helps to detect multiform track short-wave defects such as corrugation and impact of welded joint. Both the simulated and measured results indicate that the proposed approach can rapidly and effectively identify the compound track short-wave defects and possesses potential prospect of practical applications.

WBIM-GAN: A Generative Adversarial Network Based Wideband Interference Mitigation Model for Synthetic Aperture Radar

Wideband interference (WBI) can significantly reduce the image quality and interpretation accuracy of synthetic aperture radar (SAR). To eliminate the negative effects of WBI on SAR, we propose a novel end-to-end data-driven approach to mitigate WBI. Specifically, the WBI is mitigated by an explicit function called WBI mitigation–generative adversarial network (WBIM-GAN), mapping from an input WBI-corrupted echo to its properly WBI-free echo. WBIM-GAN comprises a WBI mitigation network and a target echo discriminative network. The WBI mitigation network incorporates a deep residual network to enhance the performance of WBI mitigation while addressing the issue of gradient saturation in the deeper layers. Simultaneously, the class activation mapping technique fully demonstrates that the WBI mitigation network can localize the WBI region rather than the target echo. By utilizing the PatchGAN architecture, the target echo discriminative network can capture the local texture and statistical features of target echoes, thus improving the effectiveness of WBI mitigation. Before applying the WBIM-GAN, the short-time Fourier transform (STFT) converts SAR echoes into a time–frequency domain (TFD) to better characterize WBI features. Finally, by comparing different WBI mitigation methods applied to several real measured SAR data collected by the Sentinel-1 system, the efficiency and superiority of WBIM-GAN are proved sufficiently.

Sports activity (SA) recognition based on error correcting output codes (ECOC) and convolutional neural network (CNN)

The increasing use of motion sensors is causing major changes in the process of monitoring people's activities. One of the main applications of these sensors is the detection of sports activities, for example, they can be used to monitor the condition of athletes or analyze the quality of sports training. Although the existing sensor-based activity recognition systems can recognize basic activities such as: walking, running, or sitting; they don't perform well in recognizing different types of sports activities. This article introduces a new model based on machine learning (ML) techniques to more accurately distinguish between sports and everyday activities. In the proposed method, the necessary data to detect the type of activity is collected through his two sensors: an accelerometer and a gyroscope attached to a person's foot. For this purpose, the input signals are first preprocessed and then short-time Fourier transform (STFT) is used to describe the characteristics of each signal. In the next step, each STFT matrix is used as input to a convolutional neural network (CNN). This CNN describes various motion characteristics of the sensor in the form of vectors. Finally, a classification model based on error correction output code (ECOC) is used to classify the extracted features and detect the type of SA. The performance of the proposed AS recognition method is evaluated using the DSADS database and the results are compared with previous methods. Based on the results, the proposed method can recognize sports activities with an accuracy of 99.71. Furthermore, the performance of the proposed method based on precision and recall criteria are 99.72 and 99.71, respectively, which are better than the compared methods.

Stability of the Faber-Krahn inequality for the short-time Fourier transform

We prove a sharp quantitative version of the Faber–Krahn inequality for the short-time Fourier transform (STFT). To do so, we consider a deficit δ(f;Ω)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$\\delta (f;\\Omega )$\\end{document} which measures by how much the STFT of a function f∈L2(R)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$f\\in L^{2}(\\mathbb{R})$\\end{document} fails to be optimally concentrated on an arbitrary set Ω⊂R2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$\\Omega \\subset \\mathbb{R}^{2}$\\end{document} of positive, finite measure. We then show that an optimal power of the deficit δ(f;Ω)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$\\delta (f;\\Omega )$\\end{document} controls both the L2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$L^{2}$\\end{document}-distance of f\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$f$\\end{document} to an appropriate class of Gaussians and the distance of Ω\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$\\Omega $\\end{document} to a ball, through the Fraenkel asymmetry of Ω\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$\\Omega $\\end{document}. Our proof is completely quantitative and hence all constants are explicit. We also establish suitable generalizations of this result in the higher-dimensional context.

Pattern Augmented Lightweight Convolutional Neural Network for Intrusion Detection System

As the world increasingly becomes more interconnected, the demand for safety and security is ever-increasing, particularly for industrial networks. This has prompted numerous researchers to investigate different methodologies and techniques suitable for intrusion detection systems (IDS) requirements. Over the years, many studies have proposed various solutions in this regard, including signature-based and machine learning (ML)-based systems. More recently, researchers are considering deep learning (DL)-based anomaly detection approaches. Most proposed works in this research field aim to achieve either one or a combination of high accuracy, considerably low false alarm rates (FARs), high classification specificity and detection sensitivity, lightweight DL models, or other ML and DL-related performance measurement metrics. In this study, we propose a novel method to convert a raw dataset to an image dataset to magnify patterns by utilizing the Short-Term Fourier transform (STFT). The resulting high-quality image dataset allowed us to devise an anomaly detection system for IDS using a simple lightweight convolutional neural network (CNN) that classifies denial of service and distributed denial of service. The proposed methods were evaluated using a modern dataset, CSE-CIC-IDS2018, and a legacy dataset, NSLKDD. We have also applied a combined dataset to assess the generalization of the proposed model across various datasets. Our experimental results have demonstrated that the proposed methods achieved high accuracy and considerably low FARs with high specificity and sensitivity. The resulting loss and accuracy curves have demonstrated the efficacy of our raw dataset to image dataset conversion methodology, which is evident as an excellent generalization of the proposed lightweight CNN model was observed, effectively avoiding overfitting. This holds for both the modern and legacy datasets, including their mixed versions.

Grid-tuned ensemble models for 2D spectrogram-based autism classification

BackgroundAutism spectrum disorder (ASD) is a prevalent collection of developmental disorders that affects around one in every 36 children worldwide. Since autistic children exhibit aberrant electroencephalogram (EEG) signals in comparison to typically developing (TD) children, many authors have recently proposed several machine learning (ML) and deep learning (DL) models to classify ASD patients using EEG signals. However, these models lack extensive and large-scale feature extraction and apply multiple benchmark datasets. Furthermore, most of the proposed ML models need feature engineering to extract key features for classification. MethodThis study proposed a DL-based weighted ensemble model to avoid the need for feature engineering. The proposed model is based on finding the best suitable combination of weights using a grid search strategy, treating the highest weighted model as the most accurate, and giving more weightage. The strength of the suggested technique involves extracting features from EEG recordings by transforming raw EEG into multi-channel two-dimensional time–frequency maps (2D spectrograms) using the short-time Fourier transform (STFT). Furthermore, it applies two widely-researched public EEG datasets [i.e., the King Abdulaziz University (KAU) and the University of Sheffield (TUOS) databases] for performance comparison and better generalizability. Additionally, the performance of the proposed model has been compared with other well-known DL models, and our proposed straightforward, easily implemented models surpass these models in terms of accuracy and training time. ResultsThe suggested model yields an accuracy of 94.71% for ASD classification using the KAU dataset and 96.22% accuracy for utilizing the TUOS dataset. ConclusionsEnsembled models can efficiently enhance accuracy, stability, and reproducibility. The demonstrated finding solidifies the applicability of the weighted ensemble model to classify ASD patients utilizing EEG data. The suggested model might be used as a supplemental tool to assist neurologists in diagnosing ASD.

Domain adaptation-based sparse time-frequency analysis and its application on seismic attenuation estimation

Time-frequency (TF) transform is a commonly used tool for geologic structure interpretation and attenuation estimation, mainly including the linear and nonlinear methods. The former consists of short-time Fourier transform, continuous wavelet transform, S-transform, etc., which can be efficiently implemented. However, they suffer from the Gabor uncertainty principle, which limits their TF readability. The latter can effectively enhance TF readability; nevertheless, it has several unavoidable drawbacks, such as low computational efficiency and parameter selection. For seismic sparse TF (STF) analysis (STFA), we develop a domain adaptation-based STFA (DASTFA) model. This model mainly contains two steps. First, we suggest an STF network (STFNet) for mapping a 1D seismic signal to a 2D STF spectrum, which is pretrained using synthetic traces and STF labels. Second, based on the pretrained STFNet, we design a DASTFA model for transferring to field data. Note that the synthetic traces are generated using well-log data and horizon at the study area, located in the Ordos Basin, China, which can reduce the gap between synthetic and field data. Afterward, we use synthetic traces with their STF labels for model pretraining in the first step and field data without STF labels for model transferring in the second step. Here, model transferring indicates transferring the pretrained model parameters based on field traces. Finally, to test the validity and effectiveness of DASTFA, we apply it to synthetic and field data. Moreover, we test the availability of seismic attenuation estimation by adopting the suggested model for 3D poststack field data volume, which achieved encouraging results.

Multi-modal physiological time-frequency feature extraction network for accurate sleep stage classification

Sleep stage classification is essential for clinical disease diagnosis and sleep quality assessment. Most of the existing methods for sleep stage classification are based on single-channel or single-modal signal, and extract features using a single-branch, deep convolutional network, which not only hinders the capture of the diversity features related to sleep and increase the computational cost, but also has a certain impact on the accuracy of sleep stage classification. To solve this problem, this paper proposes an end-to-end multi-modal physiological time-frequency feature extraction network (MTFF-Net) for accurate sleep stage classification. First, multi-modal physiological signal containing electroencephalogram (EEG), electrocardiogram (ECG), electrooculogram (EOG) and electromyogram (EMG) are converted into two-dimensional time-frequency images containing time-frequency features by using short time Fourier transform (STFT). Then, the time-frequency feature extraction network combining multi-scale EEG compact convolution network (Ms-EEGNet) and bidirectional gated recurrent units (Bi-GRU) network is used to obtain multi-scale spectral features related to sleep feature waveforms and time series features related to sleep stage transition. According to the American Academy of Sleep Medicine (AASM) EEG sleep stage classification criterion, the model achieved 84.3% accuracy in the five-classification task on the third subgroup of the Institute of Systems and Robotics of the University of Coimbra Sleep Dataset (ISRUC-S3), with 83.1% macro F1 score value and 79.8% Cohen's Kappa coefficient. The experimental results show that the proposed model achieves higher classification accuracy and promotes the application of deep learning algorithms in assisting clinical decision-making.

Improving piano music signal recognition through enhanced frequency domain analysis

Feature extraction is a crucial component in the analysis of piano music signals. This article introduced three methods for feature extraction based on frequency domain analysis, namely short-time Fourier transform (STFT), linear predictive cepstral coefficient (LPCC), and Mel-frequency cepstral coefficient (MFCC). An improvement was then made to the MFCC. The inverse MFCC (IMFCC) was combined with mid-frequency MFCC (MidMFCC). The Fisher criterion was used to select the 12-order parameters with the maximum Fisher ratio, which were combined into the F-MFCC feature for recognizing 88 single piano notes through a support vector machine. The results indicated that when compared with the STFT and LPCC, the MFCC exhibited superior performance in recognizing piano music signals, with an accuracy rate of 78.03 % and an F1 value of 85.92 %. Nevertheless, the proposed F-MFCC achieved a remarkable accuracy rate of 90.91 %, representing a substantial improvement by 12.88 % over the MFCC alone. These findings provide evidence for the effectiveness of the designed F-MFCC feature for piano music signal recognition as well as its potential application in practical music signal analysis.

An Integrated Bearing Fault Diagnosis Method Based on Multibranch SKNet and Enhanced Inception-ResNet-v2

Deep learning has recently received extensive attention in the field of rolling-bearing fault diagnosis owing to its powerful feature expression capability. With the help of deep learning, we can fully extract the deep features hidden in the data, significantly improving the accuracy and efficiency of fault diagnosis. Despite this progress, deep learning still faces two outstanding problems. (1) Each layer uses the same convolution kernel to extract features, making it difficult to adaptively select convolution kernels based on the features of the input image, which limits the network’s adaptability to different input features and leads to weak feature extraction. (2) Large number of parameters and long training time. To solve the above problems, this paper proposes an integrated deep neural network that combines an improved selective kernel network (SKNet) with an enhanced Inception-ResNet-v2, named SIR-CNN. First, based on the SKNet, a new three-branch SKNet was designed. Second, the new SKNet is embedded into a depthwise separable convolution network such that the model can adaptively select convolution kernels of different sizes during training. Furthermore, the convolution structure in the Inception-ResNet-v2 network was replaced by the improved depthwise separable convolution network to achieve effective feature extraction. Finally, the time-frequency maps of the raw vibration signals are obtained through short-time Fourier transform (STFT) and then sent to the proposed SIR-CNN network for experiments. The experimental results show that the proposed SIR-CNN achieves superior performance compared to other methods.