Search

• Proposed an adaptive exposure adjustment algorithm, eliminating the need for manual intervention during the calibration process, making the calibration process more robust to environmental changes. • Proposed an adaptive blur adjustment algorithm, enhancing the clarity of calibration images, addressing the issue of uncontrollable blurring in previous algorithms, and improving the accuracy of camera intrinsic parameter calibration. • Designed a semi-automatic high-precision calibration strategy that addresses significant discrepancies in calibration results obtained from different calibration processes in traditional algorithms, thus improving robustness. Camera calibration is a prerequisite for many computer vision tasks, such as visual measurement and visual localization. This paper proposes a new method for high-precision calibration of camera intrinsic parameters. By dynamically adjusting image brightness and blur to cope with changes in external lighting conditions and camera working distance, high-precision internal parameter measurement is achieved. This method solves the problems of the traditional camera intrinsic parameter calibration process being cumbersome and susceptible to changes in the placement of calibration boards and fluctuations in external lighting conditions. Firstly, an adaptive exposure adjustment algorithm is proposed to adjust the camera exposure value by statistically analyzing the average gray value of randomly selected rectangular regions on the calibration board, thereby solving the problem of the camera calibration accuracy being affected by overly bright or dark environments. Subsequently, an evaluation criterion and adaptive adjustment strategy for adjusting image blur are proposed. By extracting edge points of feature circles and obtaining the slope of the fitted curve, the relationship between edge sharpness and slope is utilized to adjust the image blur. Finally, the proposed method achieves high-precision semi-automatic calibration. Experimental results demonstrate that the proposed method has significant advantages in terms of correctness, robustness, and flexibility, with an average reprojection error of 0.0167. [ABSTRACT FROM AUTHOR]

Wi-Fi-based gesture recognition represents an emerging paradigm of human–computer interaction. While deep learning-based and model-based solutions are crucial for improving accuracy and generalization performance, they often depend heavily on extensive datasets and the widespread deployment of sensing devices. In this paper, we address the challenge involving limited sensing data and devices. We propose a signal-based gesture recognition method that leverages gesture coding based on stroke trajectory and feature representation based on dynamic phase changes. Specifically, we develop an endpoint detection algorithm to ensure precise gesture recognition. Additionally, a subcarrier selection algorithm is designed to select optimal subcarriers, capturing comprehensive gesture information. Extensive experiments are conducted to evaluate the performance. Results demonstrate an average accuracy of 93.67% for six gestures. This approach effectively mitigates the impact of location and environment factors on gesture characteristics, and reduces dependence on large quantity of samples and transceiver devices, obviating the need for model training. • An endpoint detection method is proposed to accurately extract the gesture information. • A subcarrier selection algorithm is proposed to select the subcarriers carrying a higher proportion of gesture-related information while being less susceptible to noise interference. • A gesture coding-based recognition method that leverages the stroke trajectory and unique feature representation is presented. This approach eliminates the need for model training and mitigates the dependence on quantity of samples and sensing devices. • Improve generalization performance of Wi-Fi-based gesture recognition system. [ABSTRACT FROM AUTHOR]

With the development of concrete structure health monitoring (SHM), the utilization of acoustic emission (AE) as an effective non-destructive testing method has attracted increasing attention among researchers. However, the acquisition of concrete AE signals in real-world scenarios is often limited, which poses challenges to the training as well as the practical application of deep learning (DL) models. To address this problem, a concrete acoustic emission signal augmentation method based on generative adversarial networks (GAN) is proposed in this paper. This method features a unique architecture where both the generator and discriminator are built upon a Bidirectional Long Short-Term Memory (BiLSTM) network. Additionally, the generator integrates an attention mechanism between the BiLSTM layers, enhancing its ability to capture the temporal dynamics of AE signals. To improve the local details of the generated signals, a temporal difference mean square error (TD-MSE) regularization is incorporated into the loss function. Experimental results demonstrate that the method not only substantially increases the amount of data available but also significantly enhances the model's generalization capabilities and classification accuracy. • Use BiLSTM in GAN to capture signal timing information. • Add an attention mechanism to BiLSTM layers for signal focus. • Employ temporal difference regularization for smoother signal generation. • The discussion analyzes improved model generalization from the augmented dataset. [ABSTRACT FROM AUTHOR]

• A novel water transparency calculation algorithm is introduced, employing deep neural network and image processing technology. This algorithm obviates the need for a water gauge, allowing the calculation of water transparency solely through the analysisof Secchi disk videos. Notably, it demonstrates heightened detection efficiency and success rates. • An enhanced Siamese tracker, termed SiamDCFF (Siamese tracker based on dual correlation feature fusion), is proposed for Secchi disk recognition. Distinguishing features of SiamDCFF include a dual correlation module and a feature fusion module. On the Secchi disk dataset, SiamDCFF achieves the highest success rate, outperforming other methods detailed in this paper. • The simplified 3D-ResNet [41] is leveraged to assess the state of the Secchi disk, transforming the critical position determination problem into an optimization challenge. The objective function involves maximizing the number of visible and invisible Secchi disk images on both sides of the critical position. This method facilitates adaptive critical position determination. • A novel Gated Recurrent Unit based on auto-regression (ARGRU) is proposed, taking into account the motion characteristics of the Secchi disk. ARGRU achieves a higher success detection rate for water transparency compared to the water gauge recognition method. Notably, ARGRU outperforms RNN [42], GRU, and LSTM [43] on the dataset presented in this paper. Water transparency is traditionally assessed through visual inspection using a lowered Secchi disk (SD) into the water, with the disappearance depth of the SD recorded as the measure of transparency. However, manual and visually-dependent frequency measurements of the SD render this process labor-intensive and time-consuming. This paper presents a comprehensive machine-vision-based algorithm designed for the automatic computation of water transparency using Secchi disk videos. The algorithm leverages multiple deep neural networks, including an enhanced Siamese tracking model, a 3D-ResNet, and an improved GRU network. Trained models directly identify the SD in the video, calculating water transparency by processing pixel information. Experimental results demonstrate the algorithm's efficacy in estimating water transparency across diverse natural environments, achieving commendable accuracy (MAE = 3.6 cm MSE = 21.5 cm RMSE = 4.6 cm), rapid processing speed (average 6.87 s), and robust stability. In comparison with the former water-gauge-based algorithm, our proposed algorithm exhibits heightened efficiency and a superior detection success rate. [ABSTRACT FROM AUTHOR]

Spatiotemporal signals are currently gaining wide popularity owing to their richness in terms of information content from both spatial and temporal aspects. Climatic studies involve the usage of such signals from different spatial locations to understand the climatic conditions, anomalies and for prediction of similar climatic zones. Measurement of the complex inter-relationships in the spatiotemporal processes is of vital importance in this regard. The predominant frequency components associated with annual, semi-annual, and quarterly seasonal variations exhibit interchangeability across spatial and temporal dimensions, thereby contributing significantly to the variability within the spatiotemporal region under investigation. Employing frequency domain analysis enables the examination of these crucial seasonal components, even amidst noise, provided effective frequency domain-based techniques are employed. This paper is based on the measurement of the variability of spatiotemporal climatic variables considered from different geographic zones using multivariate variational mode decomposition. The main advantage of the proposed index lies in the ability to measure the variability of the signals acquired from either spatially close or far-off locations on Earth. Also, the index harnesses the hidden frequency components in the spatiotemporal signals under test by comparing the aligned periodic components across different zones. The correctness of the proposed index is tested through exhaustive simulation studies and also against the weather measurements like the Earth skin temperature, humidity and precipitation acquired from spatially close and far-off locations over a period of 40 years. • Spatiotemporal variability index is a novel metric that quantifies the variability in spatiotemporal climatic signals. • The index considers predominant seasonal frequency component analysis across spatial and temporal dimensions. • Multivariate variational mode decomposition is used to get hidden signal components in spatiotemporal signals. • The proposed index is evaluated against simulated and on real world climatic signals. • The climatic signals tested are Earth skin temperature, specific humidity and precipitation signals. [ABSTRACT FROM AUTHOR]

[Display omitted] • A reliable fault diagnosis method based on SPV noise signal is established. • The number of sensors is drastically reduced by using the noise emission of SPVs. • The coherence analyses are conducted to validate effectiveness of the approach. • The number of workers is greatly reduced, especially in aging Chinese plants. The solenoid pulse valve (SPV) is a key component of dust collector. The frequent switch of SPV accelerates its damage, subsequently decrease operation efficiency of dust collector. The traditional manual inspection methods lack reliability and efficiency, while vibration-based condition monitoring is costly and sophisticated in acceleration sensor distribution and signal processing. This paper proposes a fault diagnosis approach based on SPV noise. The average coherence coefficient of SPV noise and vibration signals is above 0.8 and close to 1, which demonstrates that there is a close relationship between SPV noise and its structural vibration. Furthermore, the validation experiments indicate that the SPV fault can be clearly distinguished by its noise signal, confirms the effectiveness of proposed approach. By utilizing this noise monitoring method, the number of sensors can be reduced by about 95 %, the work intensity of the staff is greatly reduced, and production efficiency and safety are guaranteed. [ABSTRACT FROM AUTHOR]

• An unstressed antenna sensor is proposed for passive wireless crack sensing. • The crack width is encoded in both the amplitude and phase of the retransmission signal. • Cross polarization technique is employed in sensor's wireless interrogation to avoid signal self-jamming. To address the insufficient strain transfer ratio of monolithic antenna sensors and signal self-jamming in wireless interrogation, this paper presents a cross-polarized antenna sensor based on a U-shaped resonator for crack sensing. The sensor comprises an Rx antenna, a Tx antenna, and a sensing element. Employing cross polarization between Rx and Tx antennas mitigates ambient reflection, and the channel model of the sensing system based on cross polarization technique is demonstrated to reveal the working mechanism. Within the sensing element, a U-shaped resonator and movable shorted unit form an unstressed resonator, encoding the crack width in both the amplitude and phase in the retransmission signal, which are utilized to extract the sensor's resonant frequency. The equivalent circuit model is established to analyze the monitoring mechanism comprehensively. The operational performance of the sensor is simulated in CST and validated by experiment. Results demonstrate a good linear relationship between the sensor's resonant frequency and the crack width. [ABSTRACT FROM AUTHOR]

• You only need to train twice (DKNET) has been designed. • Multi-scale segmentation head based on dynamic kernel. • Multi-scale depth supervision loss function. • Processing the uncertainty of prediction results based on entropy theory. • High-resolution dynamic kernel defect detection model, RSU_DKNet. • Experimental results show that the model designed has the lowest algorithm complexity (0.836 MB). Deep learning usually needs a lot of marking training data, but in the industrial field, it is very difficult and costly to obtain large-scale marking data of weld defects. Therefore, this paper puts forward a new model of industrial small sample weld defect detection, YOTT (You only training twice), to solve this problem by minimizing the use of marking data. Firstly, a multi-scale segmentation head based on a dynamic kernel is designed, which can perceive defect features in advance and fuse multi-scale defect information, to enhance its ability to identify defect information. Secondly, combined with the multi-scale prediction information, a multi-scale depth supervision loss function is designed. In addition, to measure the randomness of the model output, the uncertainty processing of the prediction results based on the entropy model theory is designed, and the uncertainty quantification of the prediction output is obtained at the same time as the model. Finally, to further improve the performance on high-resolution sets, an RSU_DKNet model suitable for high-resolution data sets is designed, which can capture more context information without significantly increasing the computational cost. The experimental results on 9 public data sets show that the performance of the DKNet model designed in this paper is 100 times better than that of the existing advanced models. Moreover, the algorithm complexity of the method designed in this paper is the lowest, and the model size is only 0.836 MB, which has potential application value in solving the problem of industrial small sample defect detection. [ABSTRACT FROM AUTHOR]

• Propose a pseudo measurement update method for head pose determination. • A robust detector combined with heuristic wavelet denoising is designed. • A head pose determination method incorporating a robust strategy is proposed. The human head often shifts abruptly after a prolonged inactivity. When the head remains motionless, pseudo measurement correction approaches can increase the accuracy of head pose determination (HPD). However, existing detectors based on the general likelihood ratio test (GLRT) framework cannot identify zero angular rate events and are insensitive to rotational motion. In this paper, a heuristic wavelet denoising approach is designed to pre-process the input signals of a GLRT-based moving variance detector, thereby enhancing the precision of the detector. Furthermore, an approach for head pose determination based on the error-state Kalman filter (ESKF) is presented. Finally, simulation tests, attitude estimation studies of head motion mimicked by a turntable and actual head tracking experiment are used to validate the performance of the suggested method. The experimental results demonstrate that the suggested detector is more precise. The HPD approach described in this paper can achieve greater estimation precision than previous methods. [ABSTRACT FROM AUTHOR]

[Display omitted] • A novel focusing magnetic field excitation device for electromagnetic torque sensor. • Magnetic path analysis about focusing magnetic field distribution of C-shaped core. • Structure optimization about magnetic core shields. • Torsional angle measurement using focusing magnetic field torque sensor. To enhance accuracy and sensitivity of electromagnetic torque sensor, a highly focusing magnetic field is essential. This paper proposes a C-shaped excitation device designed to focus the magnetic field. The device is enclosed by dual layers of ferromagnetic metal shields. The C-shaped structure directs most of the magnetic field lines through its magnetic core, enhancing the focusing performance. Theoretical analysis of the device's focusing magnetic field distribution was conducted under three conditions: unshielded, with a single-layer shield, and with a dual-layer shield, using magnetic path analysis techniques. Using COMSOL 6.1 to examine various configurations, we identified the optimal structure for the device. Subsequently, a physical prototype was fabricated, and the magnetic field distribution was measured. Furthermore, an analysis was conducted on the phase differences of the induced voltage generated under its influence. The nonlinearity of the measured phase differences was4.37%. [ABSTRACT FROM AUTHOR]