Low Sampling Rate Research Articles

Deep-Learning-Based Signal Enhancement of Low-Resolution Accelerometer for Fall Detection Systems

In the last two decades, fall detection (FD) systems have been developed as a popular assistive technology. To support long-term FD services, various power-saving strategies have been implemented. Among them, a reduced sampling rate is a common approach for an energy-efficient system in the real world. However, the performance of FD systems is diminished owing to low-resolution (LR) accelerometer signals. To improve the detection accuracy with LR accelerometer signals, several technical challenges must be considered, including mismatch of effective features and the degradation effects. In this work, a deep-learning-based accelerometer signal enhancement (ASE) model is proposed as a front-end processor to help typical LR-FD systems achieve better detection performance. The proposed ASE model based on a deep denoising convolutional autoencoder architecture reconstructs high-resolution (HR) signals from the LR signals by learning the relationship between the LR and HR signals. The results show that the FD system using support vector machine (SVM) and the proposed ASE model at an extremely low sampling rate (sampling rate < 2 Hz) achieved 97.34% and 90.52% accuracies in the SisFall and FallAllD data sets, respectively, while those without ASE models only achieved 95.92% and 87.47% accuracies in the SisFall and FallAllD data sets, respectively. The results also demonstrate that the proposed ASE mode can be suitably combined with deep-learning-based FD systems.

Condition monitoring of train wheels using a cost-effective smart rail pad

This present paper focuses on the condition monitoring of train wheels utilizing an 3D printed rail pad that is embedded with widely available accelerometers and a strain gauge. This smart rail pad was used on a heavy haul railway line to monitor train wheels by identifying any wheel defects and measuring the respective wheel loads. A series of laboratory material tests were conducted on various 3D printing materials to identify the most suitable material for the smart rail pad. Dynamic and static loading tests were carried out to determine whether the 3D printed rail pads could withstand the typical forces exerted by a passing train. Field tests were done to determine the performance of the smart rail pads in operational conditions. Results indicated that the smart rail pads were able to identify 60% of the wheel flats and were able to measure the wheel loads with an average percent error of 6% by comparing it to the control measurements. In summary, the smart rail pads presented a good correlation between the measured wheel loads and the true values, while the identification of wheel flats was influenced by the low sampling rate of the system.

Knee Extensor Rate Of Torque Development Assessment Post-acl Reconstruction: Does Sampling Rate Matter?

Impaired quadriceps neuromuscular function is a hallmark of anterior cruciate ligament injury and reconstruction (ACLR). Recent investigations suggest that reduced knee extensor rate of torque development (RTD) is strongly associated with abnormal movement mechanics post-ACLR and persists at or beyond return to sport. RTD is typically quantified at high (≥1000 Hz) sampling rates (SR), which is a barrier to clinical assessment as most clinical equipment features low SR (100 Hz). PURPOSE: To determine how SR influences knee extensor RTD in collegiate athletes near return to sport post-ACLR. METHODS: 35 Division I athletes (age 20.6 ± 1.6 y, BMI 25.0 ± 3.3 kg/m2, 21 female) performed rapid maximal effort isometric knee extension contractions at 9.1 ± 2.5 months post-ACLR on an isokinetic dynamometer. Torque signals were sampled at 2000 Hz. RTD was characterized with multiple variables: from contraction onset to 50, 100, and 200 ms (RTD0-50, RTD0-100, and RTD0-200, respectively), in sequence from 50 to 100 ms and 100 to 200 ms (RTD50-100 and RTD100-200), as the slope of 20-80% of peak torque (RTD20-80), and as the maximal instantaneous slope (RTDPeak). Torque signals were then downsampled to 100 Hz and all RTD variables re-calculated. Limb Symmetry Indices (LSI) were computed for all variables at both SR. Bland-Altman analyses, intraclass correlation coefficients (ICC), and root-mean-square errors (RMSE) assessed the agreement between 2000 Hz and 100 Hz RTD variables for both involved limb RTD and between-limb LSIs. RESULTS: Knee extensor RTD0-50, RTD50-100, and RTD0-100 demonstrated the lowest levels of agreement between 100 Hz and 2000 Hz SR. Conversely, RTD0-200, RTD20-80, and RTDPeak had the highest levels of agreement.CONCLUSIONS: Involved limb RTD and RTD LSI for RTD0-200, RTD20-80, and RTDPeak can be accurately assessed with reduced sampling rates, thereby facilitating use of clinically available equipment for assessing knee extensor RTD post-ACLR.

Sensor location selection for continuous pulp digesters with delayed measurements

The state estimation and sensor placement for a continuous pulp digester with delayed measurements are investigated. The underlying model of interest is heat transfer in a pulp digester modeled by two coupled hyperbolic partial differential equations and an ordinary differential equation. Output measurements are considered with delay due to the possible low sampling rate. The Cayley–Tustin transformation is utilized to realize model time discretization in a late lumping manner which does not account for any type of spatial approximation or model reduction. The discrete Kalman filter is applied to estimate the system states using the delayed measurements. The selection of sensor location is addressed along with estimator design accounting for the delayed measurements and investigated by minimizing the variance of estimation error. The performance of the state estimator is evaluated, and the sensor placement is analyzed through simulation studies, which offers a planning view of sensor location in industrial applications.

Deblurring Ghost Imaging Reconstruction Based on Underwater Dataset Generated by Few-Shot Learning.

Underwater ghost imaging based on deep learning can effectively reduce the influence of forward scattering and back scattering of water. With the help of data-driven methods, high-quality results can be reconstructed. However, the training of the underwater ghost imaging requires enormous paired underwater datasets, which are difficult to obtain directly. Although the Cycle-GAN method solves the problem to some extent, the blurring degree of the fuzzy class of the paired underwater datasets generated by Cycle-GAN is relatively unitary. To solve this problem, a few-shot underwater image generative network method is proposed. Utilizing the proposed few-shot learning image generative method, the generated paired underwater datasets are better than those obtained by the Cycle-GAN method, especially under the condition of few real underwater datasets. In addition, to reconstruct high-quality results, an underwater deblurring ghost imaging method is proposed. The reconstruction method consists of two parts: reconstruction and deblurring. The experimental and simulation results show that the proposed reconstruction method has better performance in deblurring at a low sampling rate, compared with existing underwater ghost imaging methods based on deep learning. The proposed reconstruction method can effectively increase the clarity degree of the underwater reconstruction target at a low sampling rate and promotes the further applications of underwater ghost imaging.

Disturbance Feedforward Control of Tip-Tilt Mirror With Gyro Measuring for Large-Amplitude Vibration Rejection

A traditional tip-tilt mirror control system with an image loop is used to effectively reject structural vibrations in optical telescopes. However, its bandwidth is restricted by low sampling rate and large time delay of imaging sensors, which is insufficient to suppress disturbances outside the bandwidth, even the large-amplitude disturbances within the bandwidth. This paper proposed a disturbance feedforward method base on fusion of inertial gyro and imaging sensor to solve this problem. Accurate measurement of disturbance is the premise of disturbance feedforward. Thus we utilize gyro integral to obtain high-precision and high-bandwidth disturbance signals. The traditional model-based inverse control suffers from model accuracy, which reduces the control bandwidth and response to low-frequency disturbances. Therefore, an optimized feedforward controller is proposed to reduce the dependence on the model accuracy and break the limitation of bandwidth, which realizes the elimination of large-amplitude disturbances. Simulations and experiments demonstrate that the proposed method can not only accurately measure high-bandwidth disturbance signals, but also significantly suppress large-amplitude and narrow-band disturbances. The results suggest that gyro measurement provides great convenience for the optimized feedforward method applied to other servo control system.

AEYE: A deep learning system for video nystagmus detection.

BackgroundNystagmus identification and interpretation is challenging for non-experts who lack specific training in neuro-ophthalmology or neuro-otology. This challenge is magnified when the task is performed via telemedicine. Deep learning models have not been heavily studied in video-based eye movement detection.MethodsWe developed, trained, and validated a deep-learning system (aEYE) to classify video recordings as normal or bearing at least two consecutive beats of nystagmus. The videos were retrospectively collected from a subset of the monocular (right eye) video-oculography (VOG) recording used in the Acute Video-oculography for Vertigo in Emergency Rooms for Rapid Triage (AVERT) clinical trial (#NCT02483429). Our model was derived from a preliminary dataset representing about 10% of the total AVERT videos (n = 435). The videos were trimmed into 10-sec clips sampled at 60 Hz with a resolution of 240 × 320 pixels. We then created 8 variations of the videos by altering the sampling rates (i.e., 30 Hz and 15 Hz) and image resolution (i.e., 60 × 80 pixels and 15 × 20 pixels). The dataset was labeled as “nystagmus” or “no nystagmus” by one expert provider. We then used a filtered image-based motion classification approach to develop aEYE. The model's performance at detecting nystagmus was calculated by using the area under the receiver-operating characteristic curve (AUROC), sensitivity, specificity, and accuracy.ResultsAn ensemble between the ResNet-soft voting and the VGG-hard voting models had the best performing metrics. The AUROC, sensitivity, specificity, and accuracy were 0.86, 88.4, 74.2, and 82.7%, respectively. Our validated folds had an average AUROC, sensitivity, specificity, and accuracy of 0.86, 80.3, 80.9, and 80.4%, respectively. Models created from the compressed videos decreased in accuracy as image sampling rate decreased from 60 Hz to 15 Hz. There was only minimal change in the accuracy of nystagmus detection when decreasing image resolution and keeping sampling rate constant.ConclusionDeep learning is useful in detecting nystagmus in 60 Hz video recordings as well as videos with lower image resolutions and sampling rates, making it a potentially useful tool to aid future automated eye-movement enabled neurologic diagnosis.

Edge detection in single multimode fiber imaging based on deep learning.

We propose a new edge detection scheme based on deep learning in single multimode fiber imaging. In this scheme, we creatively design a novel neural network, whose input is a one-dimensional light intensity sequence, and the output is the edge detection result of the target. Different from the traditional scheme, we can directly obtain the edge information of unknown objects by using this neural network without rebuilding the image. Simulation and experimental results show that, compared with the traditional method, this method can get better edge details, especially in the case of low sampling rates. It can increase the structural similarity index of edge detection imaging from 0.38 to 0.62 at the sampling rate of 0.6%. At the same time, the robustness of the method to fiber bending is also proved. This scheme improves the edge detection performance of endoscopic images and provides a promising way for the practical application of multimode fiber endoscopy.

NDT Method for Line Laser Welding Based on Deep Learning and One-Dimensional Time-Series Data

Welding testing is particularly important in industrial systems, but there are still some deficiencies in terms of testing performance, anti-noise capability and defect identification in current mainstream welding non-destructive testing technologies. With the development of structured-light non-destructive testing technology, deep learning technology, signal processing technology and other fields, various possibilities have emerged that make it possible to propose new ideas for welding non-destructive testing. This study used a laser sensor to propose a non-destructive method for testing welding defects in seam contours. In order to solve the problems of low sampling rates and poor recognition accuracy in traditional methods of welding defect detection, the proposed method introduces image coding into laser sensors and applies deep-learning algorithms to the classification and detection of weld defect images. By preprocessing the weld seam by encoding one-dimensional data as two-dimensional images, this method develops a framework for the detection and classification of pre-coded laser weld seam images. After taking the original extracted weld image center trajectory data as one-dimensional sequence data, we utilized the method of encoding one-dimensional time series data as two-dimensional time-series images. In doing so, the one-dimensional laser data can be encoded into the corresponding two-dimensional images and, with the application of a deep neural network, welding defect classification and detection can be realized. Experimentation was used to verify that the proposed method is of higher accuracy than traditional methods for classifying and detecting defects directly from two-dimensional welding images.

Multi-Layer Model Enabled Fault Location for Underground Cables in MMC-HVDC Grids Considering Distributed and Frequency Dependent Line Parameters

Accurate fault location for underground cables in MMC-HVDC grids reduces the time finding faults as well as the system operational costs. For MMC-HVDC grids, the data window during faults is usually short. During such severe transients, the frequency dependent line parameters of the underground cables have strong influence on the fault location accuracy. In this paper, a novel fault location algorithm for underground cables in MMC-HVDC grids is proposed. First, a multi-layer model for the underground cable using partial differential equations is proposed, which considers the distributed and frequency dependent line parameters in time domain. Afterwards, the numerical finite difference solution of the multi-layer model is proposed to obtain the voltage distributions along the entire cable. Finally, the voltage method is applied to locate faults. Numerical experiments show that the multi-layer model accurately considers the frequency dependency of the cable parameters, and the solution of the model obtains accurate voltage distribution even during system transient. Extensive test cases prove that the proposed algorithm locates faults with less errors than the existing algorithm. The method is also compatible with a relatively low sampling rate (20 kHz) and a short data window (5 ms).

Super-Resolution Channel Sounding Method of Time Delay for Millimeter-Wave Channel

To satisfy the dynamic requirements of the millimeter-waveband, the higher time delay resolution is required. Time delay resolution of pseudo-noise (PN) cross-correlation channel sounding method can be limited by chip rate and sampling rate of analog to digital converter (A/D). In this letter, the multipath information in the chip is approximated using equivalent sequential sampling at a lower sampling rate, followed by the deconvolution method to derive the channel impulse response (CIR). Compared with the PN cross-correlation method, the time delay resolution of new method is much smaller than the chip width with a low sample rate of A/D. To improve time delay resolution over a wide range of time delays, we integrate the novel method with the PN cross-correlation method. Moreover, we have designed a novel channel simulator in order to verify the effectiveness of super-resolution approach of time delay.

Comparison of DC-Value Method and Kramers–Kronig Receiver in Optical OFDM SSB-DD Transmission

Short reach optical communications are desperate for cost-effective solutions to improve capacity and compatibility in dealing with massive access and applications. Single sideband-direct detection (SSB-DD) is highly recognized for improving capacity with single PD, but it is confronted with signal-signal beating interference (SSBI). Recently, a DC-Value method has been proposed for optical signal phase retrieval for single carrier signal transmission. To further explore the effectiveness and compatibility of DC-Value method, in this letter for the first time, the orthogonal frequency division multiplexing (OFDM)-based multi-carrier modulation formats are processed by the DC-Value method in the receiver. Simulation results demonstrate that, for both OFDM and Nyquist-shaped single carrier (SC), the DC-Value method is beneficial in applying lower sampling rate to achieve comparable transmission performance than that of Kramers-Kronig receiver. In addition, althrough OFDM holds 2 dB increment of the optimum carrier-signal power ratio (CSPR) than Nyquist-SC, it offers more robustness to fibre transmission.

Pulsed Laser Ranging Techniques Based on Digital Signal Processing Methods for Proximity Fuze

Laser fuze systems play an emerging part in military technology bases nowadays. In the paper, a solution is developed to improve the accuracy of the laser rangefinder system in laser fuze without changing the hardware design of the digital signal processor. The proposed algorithm will combine real-time pulse processing techniques used in target range measurement. The new algorithm interpolates returned data using a relatively low sampling rate to a higher one. It applies a cross-correlation technique using high-resolution reference waveform models to restore the position of the super-sampled start and return pulses. Based on the Monte Carlo method and simulating the waveform in the noise environments, 100 trials were judged on the standard deviation of the range estimations. With the presentation of detection performance comparison results between range estimation algorithms of laser fuzes, this paper can provide guidance and references for the application in the proximity fuze.

Intrusion detection technique based on flow aggregation and latent semantic analysis

Traditional network intrusion detection systems cannot identify new burgeoning invasive activities due to the inconspicuous features of malicious behaviors and the enormous increase of data transmitted via different devices. For the inconspicuous features, a novel aggregated flow-based inspection is suggested to amplify features of malicious behaviors. With regards to the enormous amount of data, a new data analysis method is introduced for efficiently classifying network traffic in this paper, which utilized the topic model to construct a doc-word matrix from statistical features and then analyzes latent semantic information to determine whether an aggregated flow is malicious. The performance of the proposed technique is evaluated using CIC-IDS2017, UNSW-NB15, and NSL-KDD datasets, with the results indicating that our technique achieves higher performance than other competing methods. Additionally, the ROC curves demonstrate that the proposed technique is capable of accurate classification even at a low sample rate.

Development of a Robust Machine Learning Model to Monitor the Operational Performance of Fixed-Post Multi-Blade Vertical Sawing Machines

Monitoring the operational performance of the sawmilling industry has become important for many applications including strategic and tactical planning. Small-scale sawmilling facilities do not hold automatic production management capabilities mainly due to using obsolete technology which is an effect of low financial capacity and focus their strategy on increasing value recovery and saving resources and energy. Based on triaxial acceleration data collected over five days at a sampling rate of 1 Hz, a robust machine learning model was developed with the purpose of using it to infer the operational events based on lower sampling rates adopted as a strategy to collect long-term data. Among its performance metrics, the model was characterized in its training phase by a very high overall classification accuracy (CA = 98.7%), F1 score (98.4%) and a very low error rate (LOG LOSS = 5.6%). For a three-class problem, it worked very well in classifying the main events related to the operation of the machine, with active work being characterized by an F1 score of 99.6% and an error of 3.6%. By accounting for the same metrics, the model was proven to be invariant to the sampling rates of up to 0.05 Hz (20 s) and produced even better results in the testing phase (CA = 98.9%, F1 = 98.6%, LOG LOSS = 5.5%, for a testing sample extracted at 0.05 Hz), while there were no differences in the share of class data irrespective of the sampling rate. The developed model not only preserves a high classification performance in the training and testing phases but it also seems to be invariant to lower sampling rates, making it useful for prediction over data collected at low sampling rates. In turn, this would enable the use of cheap data collectors to be operated for extended periods of time in various locations and will save human resources and money associated with data collection. Further tests would be required only for validation and they could be supported by collecting and feeding new data to the model to infer the long-term performance of similar sawmilling machines.

Hadamard-Haar (2D) dual domain ghost imaging

We propose a Hadamard-2D Haar transform computational ghost imaging scheme, which uses the rearranged Hadamard illumination pattern for sampling, and obtains the 2D-Haar wavelet coefficients directly by performing a low-order Hadamard transform on the acquired signal, and the information of the target object is obtained by the inverse transform. This scheme improves the problem of noise sensitivity of the computational ghost imaging based on Hadamard and the problem of weak illumination intensity in the high frequency region of 2D-Haar wavelet transform-based computational ghost imaging. Theoretical and experimental proofs show that the method is more robust to external noise and has more advantages in reconstructing images at low sampling rates thanks to the extraction property of 2D-Haar wavelet for edge information.

Interleave-sampled photoacoustic imaging: a doubled and equivalent sampling rate for high-frequency imaging.

High-frequency photoacoustic (PA) imaging (>20 MHz) requires data acquisition (DAQ) with a commensurately high sampling rate, which leads to hardware challenges and increased costs. We report here a new, to the best of our knowledge, method-interleave-sampled PA imaging-that enables high-frequency imaging with a relatively low sampling rate, e.g., a 41.67-MHz sampling rate with a 30-MHz transducer. This method harnesses two acquisitions at a low sampling rate to effectively double the sampling rate which consequently reduces the frame rate by a factor of two. It modulates the delay of the light pulses and can thus be applied to any PA DAQ system. We perform both phantom and in vivo studies with a 30-MHz transducer. The results demonstrate that interleaved sampling at 41.67 MHz can capture high frequency information above 30 MHz but a conventional 41.67-MHz sampling rate cannot. The axial and lateral resolution are as high as 63 µm and 91 µm via interleaved sampling which are much higher than those of conventional 41.67-MHz sampling (130 µm and 136 µm).

Velocity measurements in developing narrow open-channel flows with high free-stream turbulence: Acoustic Doppler Velocimetry (ADV) vs Laser Doppler Anemometry (LDA)

A developing narrow open-channel flow has been investigated using Acoustic Doppler Velocimetry (ADV) and Laser Doppler Anemometry (LDA). The objectives were to first characterize the flow environment with the LDA system alone, then quantify the intrusion effect of the ADV sensor immersion, and finally compare ADV-LDA measurements. The main features of the flow have been described. The turbulence levels measured in the outer flow region are high and almost isotropic due to the specificities of the flow (3D, narrow, developing). This contributes to a flattening of the mean streamwise velocity profile in this region. The intrusion effect of the ADV sensor is found to be dependent on Froude number (Fr=U0/gH with U0 the discharge velocity, H the flow depth, and g the gravity acceleration). Vertical flow below the sensor is amplified while the streamwise component of the flow is enhanced for “low” Fr (≤0.6) and reduced for “high” Fr (≥1.1). On the other hand, turbulence quantities are not affected by the sensor presence. Compared to the LDA, the ADV is shown to underestimate the mean flow and turbulence intensities, while not affecting Reynolds shear stress measurements. The underestimation of the turbulence intensities can be attributed to the lower sampling rate and larger sampling volume of the ADV, but the underestimations of the mean velocities are more likely linked to a constant bias that the ADV seems to have or to some type of ADV-intrinsic noise. Some implications for practical application are discussed.

A Method to Study Group Relationship of College Students Based on Predictive Social Network

With the rise of social network platforms such as WeChat, Weibo, and TikTok, social networks have developed from simple social networks to complex social networks. Researchers have gradually found that the traditional data sampling methods can no longer meet the development needs of the complex social network structure. In order to save network resources, various methods of social relationship prediction have been proposed. In this paper, we propose a BTCS algorithm based on low sampling rate under cognitive model and conduct several sets of comparison experiments under different networks and different sampling rates, and the results show that the BTCS algorithm improves the prediction accuracy and reduces the prediction time under low sampling rate. To address the problems of poor stability and slow prediction speed of random sampling prediction methods, this paper proposes a CCS algorithm in colleges and universities using the characteristics of high awareness among nodes within the same college. It can effectively combine the cognitive characteristics of the nodes with the college attributes and apply them to the relationship prediction to realize the college-oriented relationship prediction. The simulation results show that the CCS algorithm is more stable than other random sampling prediction methods. The results make full use of the cognitive characteristics and college attributes of nodes in social networks; reduce the influence of multiple factors such as response time, data packet loss, and individual behavior on relationship prediction; and improve the efficiency of college student group relationship prediction, which has certain theoretical significance and application prospects.

Adaptive compressed sensing algorithm for terahertz spectral image reconstruction based on residual learning

Terahertz time-domain spectroscopy (THz-TDS) is widely applied in the field of rapid nondestructive detection of grain owing to its low photon energy and high penetrating power. Nevertheless, terahertz imaging systems suffer from the problems of long image acquisition time and massive data processing. To mitigate these issues, this work presents an adaptive compressed sensing reconstruction algorithm for terahertz spectral images based on residual learning (ATResCS). The algorithm compresses the number of data samples, reducing the amount of data required for imaging and improving the imaging speed. Further, ATResCS reduces the time complexity by employing a convolutional neural network. The algorithm is validated by acquiring terahertz spectral image data via a THz-TDS system. ATResCS outperforms conventional algorithms regarding peak signal-to-noise ratio (PSNR) and structural similarity, significantly reducing the reconstruction time and, thus, enabling real-time reconstruction. Specifically, at low sampling rates (0.1), ATResCS retains key spectral image information. The average PSNR is 0.96 – 1.015 dB higher than that of DR2-Net, reducing the average reconstruction time by 0.1 – 0.2 s. Experiments demonstrate that ATResCS has better reconfiguration capability and lower algorithm complexity, enabling high-quality and fast reconstruction of terahertz spectral images.