Search Results (4,686)

Deep neural networks (DNNs) are highly effective across many domains but are sensitive to noisy or corrupted training data. Existing noise mitigation strategies often rely on strong assumptions about noise distributions or require costly retraining, limiting their scalability. Inspired by machine unlearning, we propose a novel framework that integrates attribution-guided data partitioning, neuron pruning, and targeted fine-tuning to enhance robustness. Our method uses gradient-based attribution to probabilistically identify clean samples without assuming specific noise characteristics. It then applies sensitivity-based neuron pruning to remove components most susceptible to noise, followed by fine-tuning on the retained high-quality subset. This approach jointly addresses data and model-level noise, offering a practical alternative to full retraining or explicit noise modeling. We evaluate our method on CIFAR-10 image classification and keyword spotting tasks under varying levels of label corruption. On CIFAR-10, our framework improves accuracy by up to 10% (F-FT vs. retrain) and reduces retraining time by 47% (L-FT vs. retrain), highlighting both accuracy and efficiency gains. These results highlight its effectiveness and efficiency in noisy settings, making it a scalable solution for robust generalization. Full article

Infrared imaging technology demonstrates significant advantages in aviation safety monitoring due to its exceptional all-weather operational capability and anti-interference characteristics, particularly in scenarios requiring real-time detection of aerial objects such as airport airspace management. However, traditional infrared target detection algorithms face critical challenges in complex sky backgrounds, including low signal-to-noise ratio (SNR), small target dimensions, and strong background clutter, leading to insufficient detection accuracy and reliability. To address these issues, this paper proposes the AFK-YOLO model based on the YOLO11 framework: it integrates an ADown downsampling module, which utilizes a dual-branch strategy combining average pooling and max pooling to effectively minimize feature information loss during spatial resolution reduction; introduces the KernelWarehouse dynamic convolution approach, which adopts kernel partitioning and a contrastive attention-based cross-layer shared kernel repository to address the challenge of linear parameter growth in conventional dynamic convolution methods; and establishes a feature decoupling pyramid network (FDPN) that replaces static feature pyramids with a dynamic multi-scale fusion architecture, utilizing parallel multi-granularity convolutions and an EMA attention mechanism to achieve adaptive feature enhancement. Experiments demonstrate that the AFK-YOLO model achieves 78.6% mAP on a self-constructed aerial infrared dataset—a 2.4 percentage point improvement over the baseline YOLO11—while meeting real-time requirements for aviation safety monitoring (416.7 FPS), reducing parameters by 6.9%, and compressing weight size by 21.8%. The results demonstrate the effectiveness of dynamic optimization methods in improving the accuracy and robustness of infrared target detection under complex aerial environments, thereby providing reliable technical support for the prevention of mid-air collisions. Full article

In order to investigate the influence of different tire textures, pavement types, and vehicle parameters on the tire/road noise level and its spectrum characteristics, 19 kinds of asphalt pavement main structures of RIOHTrack full-scale test track were tested by the close-proximity trailer (CPXT) tire/road noise detection method. Considering investigated parameters such as tire texture, vehicle speed, and trailer axle weight, and relying on multi-functional road condition rapid detection vehicle and laboratory tests to collect a variety of road surface information and material parameters, a multiple-linear-regression model of tire/road surface noise level of RIOHTrack (Research Institute of Highway Full-scale Test Track) asphalt pavement was constructed. Finally, the causes of noise level differences among different influencing factors were further analyzed through spectrum characteristics. The results show that vehicle speed is the most important factor affecting tire/road noise. The noise level of different tires varies due to different textures, but the noise level among different trailer axle weights is roughly the same. Vehicle speed (v), FWD center deflection (D₀), surface asphalt mixture air voids (VV), sensor-measured texture depth (SMTD) and international roughness index (IRI) were selected to establish the noise prediction models of different tire textures. Noise spectrum analysis shows that the spectrum of different vehicle speeds is significantly wide in the full frequency range, and the spectrum variation of differently textured tires is mainly concentrated in a certain range of the peak frequency. The noise spectrum curve of porous asphalt concrete (PAC13) is significantly lower than that of other asphalt mixtures in the full frequency range above 800Hz, indicating a greater noise reduction effect. Full article

This research provides a new systematic solution to the essential issue of self-noise interference in underwater acoustic sensing signals induced by unmanned surface vehicles (USVs) operating at sea. The self-noise pertains to the near-field interference noise generated by the growing diversity and volume of acoustic equipment utilized by USVs. The generating mechanism of self-noise is clarified, and a self-noise propagation model is developed to examine its three-dimensional coupling properties within spatiotemporal fluctuation environments in the time-frequency-space domain. On this premise, the YOLOv11 object identification framework is innovatively applied to the delay-Doppler (DD) feature maps of self-noise, thereby overcoming the constraints of traditional time-frequency spectral approaches in recognizing noise with delay spread and overlapping characteristics. A comprehensive comparison with traditional models like YOLOv8 and SSD reveals that the suggested delay-Doppler YOLO (DD-YOLO) algorithm attains an average accuracy of 87.0% in noise source identification. An enhanced denoising method, termed optimized time-frequency regularized overlapping group shrinkage (OTFROGS), is introduced, using structural sparsity alongside non-convex regularization techniques. Comparative experiments with traditional denoising methods, such as the normalized least mean square (NLMS) algorithm, wavelet threshold denoising (WTD), and the original time-frequency regularized overlapping group shrinkage (TFROGS), reveal that OTFROGS outperforms them in mitigating USV self-noise. This study offers a dependable technological approach for optimizing the performance of USV acoustic systems and proposes a theoretical framework and methodology applicable to different underwater acoustic sensing contexts. Full article

Gear transmission error (GTE) is a critical factor influencing the performance and service life of gear systems, as it directly contributes to vibration, noise generation, and premature wear. The present study introduces a combined theoretical and experimental approach to characterizing GTE in a single-stage spur gear system. A six-degree-of-freedom nonlinear dynamic model was formulated to capture coupled lateral–torsional vibrations, accounting for gear mesh stiffness, bearing and coupling characteristics, and a harmonic transmission error component representing manufacturing and assembly imperfections. Simulations and experiments were conducted under healthy and eccentricity-faulted conditions, where a controlled 890 g eccentric mass induced misalignment. Frequency domain inspection of faulty gear data showed pronounced sidebands flanking the gear mesh frequency near 200 Hz, as well as harmonics extending from 500 Hz up to 1200 Hz, in contrast with the healthy case dominated by peaks confined to 50–100 Hz. STFT analysis revealed dispersed spectral energy and localized high-intensity regions, reinforcing its role as an effective fault diagnostic tool. Experimental findings aligned with theoretical predictions, demonstrating that the integrated modelling and time–frequency framework is effective for early fault detection and performance evaluation of spur gear systems. Full article

Ionospheric scintillation represents a disturbance phenomenon induced by irregular electron density variations, predominantly occurring in equatorial, auroral, and polar regions, thereby posing significant threats to Global Navigation Satellite Systems (GNSS) performance. Polar regions in particular confront distinctive challenges, including the sparse deployment of dedicated ionospheric scintillation monitoring receiver (ISMR) equipment, the limited availability of strong scintillation samples, severely imbalanced training datasets, and the insufficient sensitivity of conventional Deep Neural Networks (DNNs) to intense scintillation events. To address these challenges, this study proposes a modeling framework that integrates residual neural networks (ResNet) with the Synthetic Minority Over-sampling Technique for Regression with Gaussian Noise (SMOGN). The proposed model incorporates multi-source disturbance features to accurately estimate phase scintillation indices (${\sigma }_{\phi }$) in polar regions. The methodology was implemented and validated across multiple polar observation stations in Canada. Shapley Additive Explanations (SHAP) interpretability analysis reveals that the rate of total electron content index (ROTI) features contribute up to 64.09% of the predictive weight. The experimental results demonstrate a substantial performance enhancement compared with conventional DNN models, with root mean square error (RMSE) values ranging from 0.0078 to 0.038 for daytime samples in 2024, and an average coefficient of determination (R²) consistently exceeding 0.89. The coefficient of determination for the Pseudo-Random Noise (PRN) path estimation results can reach 0.91. The model has good estimation results at different latitudes and is able to accurately capture the distribution characteristics of the local strong scintillation structures and their evolution patterns. Full article

This study investigates the electromagnetic and NVH characteristics of an outer-rotor surface-mounted permanent magnet synchronous generator (SPMSG) for wave energy applications, focusing on the effect of rotor eccentricity. To reflect potential fault due to manufacturing or assembly defects, a 0.5 mm rotor eccentricity was introduced in finite element method (FEM) simulations. The torque ripple waveform was analyzed using fast Fourier transform (FFT) to identify dominant harmonic components that generate unbalanced electromagnetic forces and induce structural vibration. These harmonic components were further examined under variable marine operating conditions to evaluate their impact on acoustic radiation and vibration responses. Based on the simulation and analysis results, a design-stage methodology is proposed for predicting vibration and noise by targeting critical harmonic excitations, providing practical insights for marine generator design and improving long-term operational reliability in wave energy systems. Full article

A liquid-sealed single-mode–no-core–single-mode (SNS) structure fiber temperature sensor based on polyvinyl alcohol (PVA) partial replacement coating is proposed. Using a liquid-sealed glass capillary structure, the PVA solution is introduced into the SNS structure and avoids its influence by environmental humidity. Temperature can be obtained by measuring the shift of the multimode interference spectrum, which is affected by the thermal optical effect of the PVA solution. Through theoretical simulation of the sensor, the optimal NCF fiber length and coating stripped length are obtained by comprehensively considering the transmitted loss and output spectrum signal-to-noise ratio (SNR). The optimal PVA solution concentration is selected by measuring the thermo-optic coefficient (TOC) and refractive index (RI). Based on the theoretical optimization results, a PVA solution-coated SNS fiber optic temperature sensor is experimentally fabricated, and temperature-sensing characteristics are measured within −3.6 to 73.2 °C. The experimental results show that the sensor has a high sensitivity (nm/°C, maximum is 21.713 nm/°C) and has a resolution of 10⁻³ °C. ${\lambda }_{dip}$ has a stable negative linear relationship with temperature, and the correlation coefficient of the fitting curve exceeds 95%. The temperature cycling experiment and long-term stability test show that the temperature sensor has good repeatability and stability. The experimental results also show the nonlinear relationship between the temperature measurement range and sensitivity, clarify the important factors affecting the response performance of fiber temperature sensors, and provide important reference values for optical fiber temperature sensors. Full article

Respiratory effort is a critical parameter for assessing respiratory function in various pathological conditions such as obstructive sleep apnea (OSA), as well as in patients undergoing respiratory ventilation. Currently, the gold-standard method for measuring it is esophageal pressure (Pes), which is obtrusive and uncomfortable for patients. An alternative approach is using diaphragmatic electromyography (dEMG), a non-obtrusive method that directly reflects the electrical drive triggering respiratory effort, holding potential for quantifying effort. Despite progress in this area, there is still no clear agreement on the best features for assessing respiratory effort from dEMG. This feasibility study considers several time, frequency, and statistical domain features, providing a comparative analysis to determine their performance in estimating respiratory effort. In particular, we evaluate the correlation of the different features with Pes using overnight recordings from 10 OSA patients and assess their robustness across different signal quality levels with the Kruskal–Wallis test. Our results support that time-domain dEMG features such as the filtered envelope, root mean square, and waveform length (WL) exhibit moderately strong correlations (R > 0.6) with respiratory effort. In terms of robustness to noise, the best features were WL, the area under the curve, and the slope sign change, demonstrating moderately strong to fair correlations (R > 0.5) even in low- to very low-quality signals. In contrast, features like skewness, the mean frequency, and the median frequency performed poorly (R < 0.3), regardless of signal quality, likely because they focus on overall signal characteristics rather than the dynamic and transient changes associated with respiratory effort by temporal features. These findings highlight the importance of selecting optimal features to obtain a reliable estimation of respiratory effort, providing a foundation for future research on non-intrusive methods. Full article

Distributed Acoustic Sensing (DAS) is an emerging technology in the maritime domain, enabling the use of existing fiber optic cables to detect acoustic signals in the marine environment. In this study, we present an automated signal detection and classification framework for DAS data that supports near-real-time processing. Using data from the SHEFA-2 cable between the Faroe and Shetland Islands, we develop a method to identify acoustic signals and generate both labeled and unlabeled datasets based on their spectral characteristics. Principal component analysis (PCA) is used to explore separability in the labeled data, and Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) is applied to classify unlabeled data. Experimental validation using clustering metrics shows that with the full dataset, we can achieve a Davies–Bouldin Index of 0.828, a Silhouette Score of 0.124, and a Calinski–Harabasz Index of 189.8. The clustering quality degrades significantly when more than 20% of the labeled data is excluded, highlighting the importance of maintaining sufficient labeled samples for robust classification. Our results demonstrate the potential to distinguish between signal sources such as ships, vehicles, earthquakes, and possible cable damage, offering valuable insights for maritime monitoring and security. Full article

With the development of artificial intelligence technology, intelligent fault diagnosis methods based on deep learning have received extensive attention. Among them, convolutional neural network (CNN) has been widely applied in the fault diagnosis of rolling bearings due to its strong feature extraction ability. However, traditional CNN models still have deficiencies in the extraction of early weak fault features and the suppression of high noise. In response to these problems, this paper proposes a convolutional neural network (SAWCA-net) that integrates spectrum-guided dynamic variable-width convolutional kernels and dynamic interactive time-domain–channel attention mechanisms. In this model, the spectrum-adaptive wide convolution is introduced. Combined with the time-domain and frequency-domain statistical characteristics of the input signal, the receptive field of the convolution kernel is adaptively adjusted, and the sampling position is dynamically adjusted, thereby enhancing the model’s modeling ability for periodic weak faults in complex non-stationary vibration signals and improving its anti-noise performance. Meanwhile, the dynamic time–channel attention module was designed to achieve the collaborative modeling of the time-domain periodic structure and the feature dependency between channels, improve the feature utilization efficiency, and suppress redundant interference. The experimental results show that the fault diagnosis accuracy rates of SAWCA-Net on the bearing datasets of Case Western Reserve University (CWRU) and Xi’an Jiaotong University (XJTU-SY) reach 99.15% and 99.64%, respectively, which are superior to the comparison models and have strong generalization and robustness. The visualization results of t-distributed random neighbor embedding (t-SNE) further verified its good feature separability and classification ability. Full article

In autonomous railway systems, where there is no driver acting as the primary fault detector, annoying interior noise caused by track defects can go unnoticed for long periods. One of the main contributors to this phenomenon is rail corrugation, a recurring defect that generates vibrations and acoustic emissions, directly affecting passenger comfort and accelerating infrastructure deterioration. This work presents a methodology for the automatic detection of corrugated track sections, based on the mathematical modeling of the spectral content of onboard-recorded acoustic signals. The hypothesis is that these defects produce characteristic peaks in the frequency domain, whose position depends on speed but whose wavelength remains constant. The novelty of the proposed approach lies in the formulation of two functional spectral indices—IIAPD (permissive) and EWISI (restrictive)—that combine power spectral density (PSD) and fast Fourier transform (FFT) analysis over spatial windows, incorporating adaptive frequency bands and dynamic prominence thresholds according to train speed. This enables robust detection without manual intervention or subjective interpretation. The methodology was validated under real operating conditions on a commercially operated metro line and compared with two reference techniques. The results show that the proposed approach achieved up to 19% higher diagnostic accuracy compared to the best-performing reference method, maintaining consistent detection performance across all evaluated speeds. These results demonstrate the robustness and applicability of the method for integration into autonomous trains as an onboard diagnostic system, enabling reliable, continuous monitoring of rail corrugation severity using reproducible mathematical metrics. Full article

Optical frequency domain reflectometry (OFDR) is one of the key diagnostic tools for fiber optic components and circuits built on them. A low signal-to-noise ratio, resulting from the low intensity of backscattered signals, prevents the correct quantitative description of the medium parameters. Known methods of signal denoising, such as empirical mode decomposition, frequency filtering, and activation function dynamic averaging, make the signal smoother but introduce errors into its dynamic characteristics, changing the intensity of reflection peaks and distorting the backscattering level. We propose a method to reduce OFDR trace noise using elliptical arc fitting (EAF). The obtained results indicate that this algorithm efficiently processes both areas with and without contrasting back reflections, with zero distortion of Fresnel reflection peaks, and with zero attenuation error in regions without Fresnel reflections. At the same time, other methods distort reflection peaks by 14.2–42.6% and shift the correct level of Rayleigh scattering by 27.2–67.3%. Further work will be aimed at increasing the accuracy of the method and testing it with other types of data. Full article

During the cooling phase of injection molding, the conformal cooling channel system optimizes the uniformity of mold temperature, diminishes warping deformation, and contributes substantially to heightened product precision. The injection molding process involves complex process parameters that may result in uneven cooling between components, leading to prolonged cycle times, increased shrinkage depth, and warping deformation of the plastic parts. These manifestations negatively impact the surface quality and structural strength of the final product. This article combined theoretical algorithms with finite element simulation (CAE) methods to optimize complex injection molding processes. Firstly, the characteristics of six different types of materials were examined. Melt temperature, mold opening time, injection time, holding time, holding pressure, and mold temperature were chosen as optimization variables. Meanwhile, the warpage deformation and shrinkage depth of the formed sample were selected as optimization objectives. Secondly, an L27 orthogonal experimental design (OED) was established, and the signal-to-noise ratio was processed. The entropy weight method (EWE) was used to calculate the weights of the total warpage deformation and shrinkage depth, thereby obtaining the grey correlation degree. The influence of process parameters on quality indicators was analyzed using grey relational analysis (GRA) to calculate the range. A second-order polynomial regression model was established using response surface methodology (RSM) to investigate the effects of six factors on the warpage deformation and shrinkage depth of injection molded parts. Finally, a comprehensive comparison was made on the impact of various optimization methods and models on the forming parameters. Analyze according to different optimization principles to obtain the corresponding optimal process parameters. The research results indicate that under the principle of prioritizing warpage deformation, the effectiveness ranking of the three optimization analyses is RSM > OED > GRA. The minimum deformation rate is 0.1592 mm, which is 27.37% lower than before optimization. Under the principle of prioritizing indentation depth, the effectiveness ranking of the three optimization analyses is OED > GRA > RSM. The minimum depth of shrinkage is 0.0312 mm, which is 47.21% lower than before optimization. This discovery provides strong support for the optimal combination of process parameters suitable for production and processing. Full article

Underwater acoustic source localization is formulated as a feature learning problem within a machine learning framework, where a data-driven approach directly extracts source distance features from hydroacoustic signals. This study systematically compares the localization performance of four machine learning models—decision tree (DT), random forest (RF), support vector machine (SVM), and feedforward neural network (FNN) models—in both classification and regression tasks. Experimental results demonstrate that, in classification tasks, all algorithms achieve effective localization under high signal-to-noise ratio (SNR) conditions, while the DT model exhibits significant noise sensitivity in low-SNR scenarios; regression tasks show reduced model convergence overall, with only the SVM and RF models maintaining basic localization capabilities at a high SNR. For two-dimensional localization, machine learning classification algorithms are employed, revealing systematic accuracy degradation compared to one-dimensional scenarios, where only the RF and SVM models demonstrate practical value under high-SNR conditions. Validation using measured data from the SWellEx-96 experiment’s S5 event confirms that when constructing datasets with frequency-domain acoustic pressure features from the final 35 min segment, the classification task-driven DT, RF, and SVM models all demonstrate reliable localization performance, benefiting from the inherent high-SNR characteristics of the data. Full article