Switched reluctance motors (SRMs) are recognized for their robustness, simplicity, and cost-effectiveness, yet they face inherent challenges in precise speed regulation and torque ripple minimization due to their doubly salient structure and nonlinear magnetic characteristics. This study introduces a novel and high-performance FOPI(1+PID n ) multi-stage controller, specifically designed to address these critical limitations. The proposed control strategy synergistically integrates a fractional order proportional integral (FOPI) controller in the primary stage to enhance stability and transient response, with a proportional integral derivative controller augmented with a derivative filter (PID n ) in the subsequent stage to improve dynamic adaptability and noise rejection. To achieve optimal tuning of controller parameters, a particle swarm optimization with time-varying acceleration coefficients (PSO-TVAC) is employed, using a robust cost function that simultaneously minimizes speed and current errors over simulation time. The effectiveness of the FOPI(1+PID n ) controller is rigorously validated through extensive simulations in MATLAB/Simulink under a range of adverse conditions: no-load, static and dynamic loads, time-delay-induced nonlinearity, stepwise load increments, and parameter uncertainties for sensitivity analysis. Comparative assessments against conventional PID, PID n , and FOPID controllers demonstrate that the proposed controller consistently achieves superior performance, with an average improvement exceeding 45% across key dynamic response metrics including rise time, settling time, overshoot, and torque ripple. Notably, the proposed controller exhibits a 57% reduction in the cost function, a 65% reduction in integral of squared error (ISE), and enhanced robustness under uncertainty, establishing its efficacy for real-time control applications in SRMs. This work advances the state-of-the-art by offering a simple yet powerful control architecture that fuses classical and modern optimization techniques to deliver ultra-fast speed tracking, significant torque ripple attenuation, and strong resilience against nonlinear disturbances, thus broadening the operational viability of SRMs in precision-demanding industrial applications. • A novel multi-stage FOPI(1+PID n ) controller is proposed for precise SRM speed regulation and torque ripple reduction. • Combines fractional-order PI and filtered PID for robust, smooth transient control under nonlinearities and delays. • PSO-TVAC optimization automatically tunes controller parameters, ensuring fast convergence and robust performance. • MATLAB/Simulink tests show 57% lower cost function, 65% lower ISE, and 45% faster response than conventional controllers. • The FOPI(1+PID n ) controller provides efficient, scalable, real-time SRM control and supports future hardware implementation.

Session-based recommender systems face significant challenges in accurately predicting user preferences due to the limited availability of long-term historical interactions. While recent advances in deep learning and graph-based approaches have improved recommendation performance, the temporal aspects of user interactions remain underutilized. This paper identifies three critical temporal challenges in session-based recommendations: interest shifts indicated by long intervals between interactions, interaction noise from brief engagements, and system popularity effects during high-traffic periods. To address these challenges, we propose a novel Dual-channel Time-aware Graph Attention Network (DT-GAT) to incorporate temporal signal, i.e., time intervals between interactions and time differences between sessions, into session representations from both item and session perspectives. The item-wise learning channel employs a temporal graph attention network to capture interest shifts and filter interaction noise, while the session-wise learning channel utilizes a temporal graph attention network to handle inconsistent popularity trends. Additionally, we introduce a multi-temporal window processing mechanism to construct robust session representations that effectively capture short-term interests while filtering noise. Extensive experiments conducted on three real-world datasets demonstrate that DT-GAT consistently outperforms state-of-the-art baseline models. Our code is available at: https://github.com/downw/DT-GAT • We propose DT-GAT to integrate item- and session-level temporal signals. • Dual temporal GATs capture dependencies via temporal intra- and inter-session graphs. • Contrastive learning aligns dual channels to enhance session representations. • Experiments on three datasets validate the effectiveness of DT-GAT.

Direct infusion-based single-cell metabolomics analysis has the potential to isolate the causes of drug resistance and cancer progression; however, processing and analyzing the data generated remains a challenge. While many packages exist for metabolomics analysis, they are not optimized for direct infusion-based single-cell measurements, which do not rely on chromatographic separation and are typically noisier than traditional population-level methods. To address this gap, the MeDUSA (Metabolomics of direct-infusion untargeted single-cell analysis) R package was developed. MeDUSA was built especially for direct infusion-based single-cell metabolomics, with modularity, noise filtering, and user-customization in mind. In this work, we introduce the package, how to use it, and implement it in a single-cell metabolomics experiment to identify the differences between two cell lines. MeDUSA compromises several functions that deal with file import, peak picking, spectral processing, statistical analysis, and feature annotation. Each function is defined with the purpose, usage, parameters, default values, and output of an example data set. MeDUSA was built to be a modular platform that aims to be a foundation to be built upon with additional modules for the single-cell metabolomics field.

Reconfigurable intelligent surface sensors (RISs) have emerged as a promising technology for enhancing wireless indoor localization by intelligently controlling signal propagation; however, extracting reliable localization fingerprints from RIS-assisted signals remains challenging due to multipath fading, environmental noise, and nonlinear spatial–temporal channel dynamics. To address this, we propose an Adaptive Dual-Reinforcement Learning-Hybrid Spatial–Temporal Network (ADRL-HSTNet) for RIS-assisted indoor localization. The framework utilizes dual-channel RSSI and phase measurements, followed by noise filtering, normalization, and sliding-window segmentation prior to feature extraction. It then constructs enhanced representations through handcrafted feature extraction and multi-branch processing, including patch-based features, wavelet-domain representations, statistical descriptors, and multi-level segmentation masks. These heterogeneous inputs are encoded using lightweight transformer-based encoders to capture multiscale dependencies. A first reinforcement learning selector adaptively weights the most informative feature branches to produce a fused representation, which is further processed by spatial and temporal transformer modules. Their outputs are adaptively combined via a second reinforcement learning selector to obtain robust localization embedding. The model jointly performs classification, coordinate regression, and uncertainty estimation end-to-end. Experimental results across multiple RIS configurations outperformed the KAN, LSTM-KAN, and RHL-Net (compared against the proposed ADRL-HSTNet) baselines, achieving accuracies of 83.33%, 75.22%, 93.33%, and 88.89%, confirming the effectiveness of the proposed approach.

Accurate wind power prediction during ramp events remains challenging due to wind speed volatility. This study proposes a hybrid forecasting framework combining improved variational mode decomposition (VMD), a novel ramp factor (RF), and the Informer model. First, a dynamic adaptive VMD method is employed to filter noise and identify abrupt wind speed changes. Subsequently, a similar period matching algorithm, enhanced by the RF and wind speed similarity coefficients, captures historical convergence features. Finally, the Informer network fuses these features with NWP data. Experimental results demonstrate that the proposed method significantly outperforms existing models in accuracy during ramp events, enhancing grid stability.

The effect of excessive noise rejection, noise filtering and twitch threshold on mechanomyograph twitch measurements.

Marine oil film detection method based on growing hierarchical neural gas network and multi-scale threshold segmentation.

Robust student’s t mixture distribution-based Kalman filter for non-stationary heavy-tailed noise

Comparison of square and disc geometries for glass RPCs

The imaging technique called optical coherence tomography angiography (OCTA) has been used extensively in ophthalmology to identify eye conditions such as age-related macular degeneration, vascular occlusion or diabetic retinopathy. However, the multi-scale vascular architecture and noise from low image quality and eye diseases make it difficult to precisely segment the vasculature. In order to accurately segment the vasculature in OCTA, we introduced HV-OCTAMamba, a novel U-shaped network based on the Vision Mamba architecture. Inspired from the state-of-the-art models OCTAMamba and H-vmunet, HV-OCTAMamba integrates a Multi-Stream Efficient Embedding Module to extract local features, a Multi-Scale Dilated Asymmetric Convolution Module for multi-scale vasculature capturing, a Feature Recalibration and Filtering Module to filter noise and highlight target areas. The core component, the High-Order Visual State Space (H-VSS), improves feature consistency by modeling long-range dependencies through structured two-dimensional state-space (SS2D) operations. Our approach is appropriate for low-computation medical applications since it efficiently extracts the global and local features while preserving linear complexity. Extensive tests on the OCTA 3M, OCTA 6M, and ROSSA datasets showed that HV-OCTAMamba performs better than the most advanced methods in the state-of-the-art, offering a new benchmark for effective OCTA segmentation. Notably, HV-OCTAMamba achieved Dice coefficients of 87.45%, 83.18%, and 90.15% on the OCTA 3M, OCTA 6M, and ROSSA datasets, respectively. You may get the code at GitHub. 1 1 Code available at https://github.com/acvai/HV-OCTAMamba/ . • Proposing HV-OCTAMamba architecture, a novel lightweight U-shaped model for OCTA vessel segmentation. • Exploiting high-order visual state-space (H-VSS) for global context modeling. • Leveraging three new modules to enhance feature extraction and noise suppression. • Achieving consistent State-Of-The-Art performance on OCTA 3M, OCTA 6M, and ROSSA datasets. • Demonstrating computational efficiency suitable for real-world clinical settings.

Stock market prediction has long been a challenging task due to its highly volatile, nonlinear, and dynamic nature influenced by economic indicators, market sentiment, and global events. This study presents the design of an advanced deep learning model for predicting stock market growth by integrating multiple neural network architectures to capture both temporal dependencies and complex feature interactions. The proposed framework combines Long Short-Term Memory (LSTM) networks for time-series forecasting, Convolutional Neural Networks (CNN) for feature extraction, and attention mechanisms to enhance pattern recognition in historical price data and technical indicators. The model incorporates preprocessing techniques such as data normalization, noise filtering, and feature engineering to improve predictive accuracy and stability. Additionally, sentiment analysis from financial news and social media data is integrated to enrich the model’s contextual understanding of market movements. The performance of the proposed model is evaluated using standard metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and directional accuracy. Experimental results demonstrate that the advanced hybrid deep learning approach outperforms traditional statistical models and standalone machine learning techniques in predicting stock market growth trends. The study highlights the potential of deep learningdriven financial forecasting systems to assist investors, analysts, and financial institutions in making informed and data-driven investment decisions.

HighlightsA novel adaptive forgetting factor improves model accuracy during load changes.An error function adjusts filter noise in real time for stable estimation.The combined framework enhances accuracy and robustness in dynamic conditions.

Sub-pixel matching of multimodal optical images is a critical step in the combined application of multiple sensors. However, structural noise and inconsistencies arising from variations in multimodal image responses usually limit the accuracy of matching. Phase congruency mutual-structure weighted least absolute deviation (PCWLAD) is developed as a coarse-to-fine framework. In the coarse matching stage, we preserve the complete structure and use an enhanced cross-modal similarity criterion to mitigate structural information loss by phase congruency (PC) noise filtering. In the fine matching stage, a mutual-structure filtering and weighted least absolute deviation-based method is introduced to enhance inter-modal structural consistency and to accurately estimate sub-pixel displacements adaptively. Experiments on three multimodal datasets—Landsat visible-infrared, short-range visible-near-infrared, and unmanned aerial vehicle (UAV) optical image pairs—show that PCWLAD achieves superior average performance compared with eight state-of-the-art methods, attaining an average matching accuracy of approximately 0.4 pixels.

The dynamic terminal to technology and human perception, creativity and human expression, an emotion-sensitive interactive art is where creative art is capable of responding to the emotional state of the involved participants. The paper will give an elaborate layout of the evolution of adaptive art systems through the multimodal biometric sensor and intelligent computing model. Physiological stimuli to acquire real-time emotional stimuli are electroencephalography (EEG), electrocardiography (ECG), galvanic skin response (GSR) and eye-tracking measurements. Noise filtering, normalization and feature extraction are some of the advanced signal preprocessing algorithms that ensure high data representation. The machine learning and deep learning models that have been applied to classify emotional states with high precision are referred to as Emotional state SVM, Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Transformer-based models. The suggested system includes a dynamic computing system that is guided by the reinforcement learning, creating the possibility of active dialogue between the user emotions and artistic products. The system responses are constantly upgraded by the feedback control mechanisms, making them more engaging and individual. The proposed approach can be proved effective as experimental assessments show an increased rate of emotion recognition and responsiveness.

One of the main residual limitations of inertial sensors based on atom interferometry stems from laser beam distortions, which cause parasitic phase shifts and non-homogeneous matter-light couplings. Here we present numerical simulations, accompanied by analytical calculations, which quantify the impact of these effects in a cold atom gradiometer. We demonstrate that the propagation of interferometric laser beam aberrations, combined with initial asymmetry and significant time-of-flight expansion of the two atomic sources, limit the common-mode rejection of phase noise in a differential configuration. The resulting deviations in gravitational acceleration and its gradient are within reach of current experimental devices. Our study allows us to evaluate the surface quality requirements for retroreflective optics in cold-atom gradiometers of various baselines, and can be extended to other sensors based on different interferometer geometries.

Abstract Parkinson's disease (PD) is a neurodegenerative disease, characterized by the loss of dopaminergic neurons in the substantia nigra of the brain, and resulting in motor impairments, such as tremor or rigidity. It has been difficult to make an early and accurate diagnosis because of overlapping symptoms with other neurological diseases and limitations in conventional diagnostic tools. In order to tackle the challenges in PD, we propose a new hybrid architecture of SRGAN and SCO. Unlike previous approach for Alzheimer's our method enhances low-resolution MRI and PET images presenting higher diagnostic potential. The approach could be carried out in clinics with limited screening capability. Super-Resolution GAN (SRGAN) An SRGAN was constructed which used a customised variation of DeepResNet (i.e ResNet-150 with skip connections), the Generative Adversarial Network (GAN) and the Gaussian filter to enhance the quality of the MRI and PET images. Various operations were performed on the input images as pre-processing, such as, contrast stretching, noise filtering, gamma correction and pixel-normalization to clear input for the model. Our SRGAN model was better performing than the existing CNN models like DenseNet, AlexNet, VGG19 and ResNet with an accuracy of 96.5%, F1 score of 94.5% and balanced accuracy (BAAC) of 93.5%. Moreover, the Skip Connection Optimization (SCO) of the SRGAN performed better than some optimization algorithms including Genetic Algorithms (GA), Ant Colony Optimization (ACO), Grey Wolf Optimizer (GWO), Stochastic Gradient Descent (SGD), ADAM, and AdaMax. The results demonstrate that the proposed model outperforms existing methods for the prediction of Parkinson's disease.

Robust grounding and shielding are critical to ensure the required detector performance of the upgraded tracking detector of the ATLAS experiment at the HL-LHC. This report presents the grounding and shielding strategy developed to avoid ground loops, enhance common-mode noise rejection, and maintain shielding integrity for the silicon pixel modules of the so-called Inner Tracker. Results from electromagnetic compatibility testing of the first multi-module structure are reported. Noise sensitivity to injected electric and magnetic fields under realistic conditions is quantified. Furthermore, the grounding and shielding verification method and overall strategy for detector integration, including the use of the so-called Ground Fault Monitor system, are discussed.

ABSTRACT The flow rates of the air and hydrogen and the gas pressures of the cathode and anode have a significant effect on the performance of the proton exchange membrane fuel cell (PEMFC). In order to improve the robustness of the PEMFC gas supply control system and thus its ability to regulate the gas flow rates and the gas pressures, firstly, two observers, whose error feedback terms are constructed using a neural network, are designed to estimate the gas pressures within the stack and thus to estimate the excess ratio of oxygen and hydrogen. Secondly, four extended observers are used for the feedback linearization of the gas supply system, and the sliding mode control laws based on the Hamilton‐Jacobi inequality are designed for fast and accurate adjustment of the air and hydrogen flow rates supplied to the stack and the gas pressures of the two poles within the stack. Simulation results indicate that the neural network observer has a stronger pressure estimation capability and noise rejection ability than the linear extended observer, and the proposed controllers have minimal overshoot and settling times for the regulation of air and hydrogen flow rates and pressures compared to the proportional‐integral‐derivative controllers or the feedback linearization controllers.

To address the issue of misjudgment in traditional connected domain marking algorithms during the dense assembly and welding of SMT components, a method combining connected domain marking with localized watershed algorithms has been proposed. After preprocessing the component images, including grayscale conversion and noise filtering, the connected domain marking algorithm’s identified soldered areas are masked and processed. The soldered areas are then extracted, histogram equalization and distance transformation are performed, and the watershed algorithm is used to segment the misjudged soldering areas. An SMT solder defect detection test bench was developed on the LabVIEW platform, and a comparative experiment was conducted with traditional connected domain algorithms. The experimental results demonstrate that the optimized algorithm not only maintains the accuracy of traditional connected-component labeling but also significantly reduces the false-positive rate in densely soldered environments. Consequently, it exhibits stronger adaptability and robustness to interference, better fulfilling the requirements of real-world engineering applications. This method exhibits superior environmental adaptability and markedly higher interference resistance, rendering it well-suited to real-world engineering applications.

Stock market prediction has long been a challenging task due to its highly volatile, nonlinear, and dynamic nature influenced by economic indicators, market sentiment, and global events. This study presents the design of an advanced deep learning model for predicting stock market growth by integrating multiple neural network architectures to capture both temporal dependencies and complex feature interactions. The proposed framework combines Long Short-Term Memory (LSTM) networks for timeseries forecasting, Convolutional Neural Networks (CNN) for feature extraction, and attention mechanisms to enhance pattern recognition in historical price data and technical indicators. The model incorporates preprocessing techniques such as data normalization, noise filtering, and feature engineering to improve predictive accuracy and stability. Additionally, sentiment analysis from financial news and social media data is integrated to enrich the model’s contextual understanding of market movements. The performance of the proposed model is evaluated using standard metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and directional accuracy. Experimental results demonstrate that the advanced hybrid deep learning approach outperforms traditional statistical models and standalone machine learning techniques in predicting stock market growth trends. The study highlights the potential of deep learning-driven financial forecasting systems to assist investors, analysts, and financial institutions in making informed and data-driven investment decisions.