Search Results (22)

High-frequency (HF) injection is a widely used technique for low-speed implementation of position sensorless permanent magnet synchronous motor control. A key component of this technique is the tracking loop control system, which extracts rotor position error and utilizes proportional–integral regulation as a position observer for estimating the rotor position. Generally, this process relies on band-pass filters (BPFs) and low-pass filters (LPFs) to modulate signals in the quadrature current to obtain rotor position error information. However, limitations in filter accuracy and dynamic response lead to prolonged convergence times and timing inconsistencies in the estimation process, which affects real-time motor control performance. To address these issues, this study proposes an exponential moving average (EMA)-based scheme for rotor position error extraction, offering a rapid response under dynamic conditions such as direction reversals, step speed changes, and varying loads. EMA is used to pass the original rotor position information carried by the quadrature current signal, which contains HF components, with a specified smoothing factor. Then, after the synchronous demodulation process, EMA is employed to extract rotor position error information for the position observer to estimate the rotor position. Due to its computational simplicity and fast response in handling dynamic conditions, the proposed method can serve as an alternative to BPF and LPF, which are commonly used for rotor position information extraction, while also reducing computational burden and improving performance. Finally, to demonstrate its feasibility and effectiveness in improving rotor position estimation accuracy, the proposed system is experimentally validated by comparing it with a conventional system. Full article

The accurate segmentation of medical images is of great importance in many clinical applications and is generally achieved by training deep learning networks on a large number of labeled images. However, it is very hard to obtain enough labeled images. In this paper, we develop a novel semi-supervised segmentation method (called PE-MT) based on the uncertainty-aware mean teacher (UA-MT) framework by introducing a perturbation-enhanced exponential moving average (pEMA) and a residual-guided uncertainty map (RUM) to enhance the performance the student and teacher models. The former is used to alleviate the coupling effect between student and teacher models in the UA-MT by adding different weight perturbations to them, and the latter can accurately locate image regions with high uncertainty via a unique quantitative formula and then highlight these regions effectively in image segmentation. We evaluated the developed method by extracting four different cardiac regions from the public LASC and ACDC datasets. The experimental results showed that our developed method achieved an average Dice similarity coefficient (DSC) of 0.6252 and 0.7836 for four object regions when trained on 5% and 10% labeled images, respectively. It outperformed the UA-MT and can compete with several existing semi-supervised learning methods (e.g., SASSNet and DTC). Full article

Recognizing dangerous driving actions is critical for improving road safety in modern transportation systems. Traditional Driver Monitoring Systems (DMSs) often face challenges in terms of lightweight design, real-time performance, and robustness, especially when deployed on resource-constrained embedded devices. This paper proposes a novel method based on 3D skeletal data, combining Graph Spatio-Temporal Feature Representation (GSFR) with a Broad Learning System (BLS) to overcome these challenges. The GSFR method dynamically selects the most relevant keypoints from 3D skeletal data, improving robustness and reducing computational complexity by focusing on essential driver movements. The BLS model, optimized with sparse feature selection and Principal Component Analysis (PCA), ensures efficient processing and real-time performance. Additionally, a dual smoothing strategy, consisting of sliding window smoothing and an Exponential Moving Average (EMA), stabilizes predictions and reduces sensitivity to noise. Extensive experiments on multiple public datasets demonstrate that the GSFR-BLS model outperforms existing methods in terms of accuracy, efficiency, and robustness, making it a suitable candidate for practical deployment in embedded DMS applications. Full article

Unsupervised domain adaptation (UDA) enables training a model on labeled source data to perform well in a target domain without supervision, which is especially valuable in vision-based semantic segmentation. However, existing UDA methods often struggle with accurate semantic labeling at object boundaries and recognizing minor categories in the target domain. This paper introduces a novel UDA framework—SamDA—that incorporates the Segment Anything Model (SAM), a large-scale foundational vision model, as the mask generator to enhance edge segmentation performance. The framework comprises three core modules: a cross-domain image mixing module, a self-training module with a teacher–student network, and exponential moving average (EMA). It also includes a finetuning module that leverages SAM-generated masks for pseudo-label matching. Evaluations on the GTA5 and Cityscapes datasets demonstrate that SamDA achieves a mean IoU (mIoU) of 75.2, surpassing state-of-the-art methods such as MIC-DAFormer by 1.0 mIoU and outperforming all ResNet-based approaches by at least 15 mIoU. Moreover, SamDA significantly enhances the segmentation of small objects like bicycles, riders, and fences, with, respective, IoU improvements of 4.5, 5.2, and 3.8 compared to baseline models. Full article

The Border Gateway Protocol (BGP) is the backbone of inter-domain routing on the internet, but its susceptibility to both benign and malicious anomalies creates substantial risks to both network reliability and security. In this study, we present a new approach for deep learning-based BGP anomaly detection utilizing duplicate announcements, which are known to be a symptom of routing disruptions. We developed our methodology based on public BGP data from RIPE and Route Views. We used the number of duplicate announcements as a baseline against which we checked for sporadic and time-based anomalies. Here, we propose a deep learning framework based on the Exponential Moving Average (EMA) model in combination with Autoencoder for anomaly identification. We also apply the Temporal-oriented Synthetic Minority Over-Sampling Technique (T-SMOTE) to overcome data imbalance. Comparative evaluations show that the Autoencoder model is significantly better than LSTM and that existing state-of-the-art methods have higher accuracy, precision, recall, and F1 scores. This study proposes a reliable, scalable, and rapid framework for real-time BGP adversary detection, which improves network security and resilience. Full article

Due to the labor-intensive manual annotations for nuclei segmentation, point-supervised segmentation based on nuclei coordinate supervision has gained recognition in recent years. Despite great progress, two challenges hinder the performance of weakly supervised nuclei segmentation methods: (1) The stable and effective segmentation of adjacent cell nuclei remains an unresolved challenge. (2) Existing approaches rely solely on initial pseudo-labels generated from point annotations for training, and inaccurate labels may lead the model to assimilate a considerable amount of noise information, thereby diminishing performance. To address these issues, we propose a method based on center-point prediction and pseudo-label updating for precise nuclei segmentation. First, we devise a Gaussian kernel mechanism that employs multi-scale Gaussian masks for multi-branch center-point prediction. The generated center points are utilized by the segmentation module to facilitate the effective separation of adjacent nuclei. Next, we introduce a point-guided attention mechanism that concentrates the segmentation module’s attention around authentic point labels, reducing the noise impact caused by pseudo-labels. Finally, a label updating mechanism based on the exponential moving average (EMA) and k-means clustering is introduced to enhance the quality of pseudo-labels. The experimental results on three public datasets demonstrate that our approach has achieved state-of-the-art performance across multiple metrics. This method can significantly reduce annotation costs and reliance on clinical experts, facilitating large-scale dataset training and promoting the adoption of automated analysis in clinical applications. Full article

Makeup modifies facial textures and colors, impacting the precision of face anti-spoofing systems. Many individuals opt for light makeup in their daily lives, which generally does not hinder face identity recognition. However, current research in face anti-spoofing often neglects the influence of light makeup on facial feature recognition, notably the absence of publicly accessible datasets featuring light makeup faces. If these instances are incorrectly flagged as fraudulent by face anti-spoofing systems, it could lead to user inconvenience. In response, we develop a face anti-spoofing database that includes light makeup faces and establishes a criterion for determining light makeup to select appropriate data. Building on this foundation, we assess multiple established face anti-spoofing algorithms using the newly created database. Our findings reveal that the majority of these algorithms experience a decrease in performance when faced with light makeup faces. Consequently, this paper introduces a general face anti-spoofing algorithm specifically designed for light makeup faces, which includes a makeup augmentation module, a batch channel normalization module, a backbone network updated via the Exponential Moving Average (EMA) method, an asymmetric virtual triplet loss module, and a nearest neighbor supervised contrastive module. The experimental outcomes confirm that the proposed algorithm exhibits superior detection capabilities when handling light makeup faces. Full article

This study evaluates machine learning models for stock market prediction in the European stock market EU50, with emphasis on the integration of key technical indicators. Advanced techniques, such as ANNs, CNNs and LSTMs, are applied to analyze a large EU50 dataset. Key indicators, such as the simple moving average (SMA), exponential moving average (EMA), moving average convergence/divergence (MACD), stochastic oscillator, relative strength index (RSI) and accumulation/distribution (A/D), were employed to improve the model’s responsiveness to market trends and momentum shifts. The results show that CNN models can effectively capture localized price patterns, while LSTM models excel in identifying long-term dependencies, which is beneficial for understanding market volatility. ANN models provide reliable benchmark predictions. Among the models, CNN with RSI obtained the best results, with an RMSE of 0.0263, an MAE of 0.0186 and an R² of 0.9825, demonstrating high accuracy in price prediction. The integration of indicators such as SMA and EMA improves trend detection, while MACD and RSI increase the sensitivity to momentum, which is essential for identifying buy and sell signals. This research demonstrates the potential of machine learning models for refined stock prediction and informs data-driven investment strategies, with CNN and LSTM models being particularly well suited for dynamic price prediction. Full article

To improve the dynamic and steady-state control performance of permanent magnet synchronous motors under the three-vector model predictive current control method, this study proposes a switching current predictive control method based on the exponential moving average algorithm, which evaluates the magnitude of the change of the q-axis current slope in real time to discriminate the motor’s operating conditions and selects the optimal control method for different operating conditions. Meanwhile, the traditional three-vector model predictive current control method is improved by introducing a comparison mechanism for the q-axis current slope to select the second effective voltage vector, avoiding the secondary optimization calculation of the value function and reducing the computational complexity of the traditional method. By comparing the proposed method with the traditional three-vector model predictive current control method, the experimental results prove that the proposed method improves the system’s dynamic response and steady-state performance. Full article

With the popularity of social media, online opinion analysis is becoming more and more widely and deeply used in management studies. Automatically recognizing the sentiment of user reviews is a crucial tool for opinion analysis research. However, previous studies mainly have focused on specific scenarios or algorithms that cannot be directly applied to real-world opinion analysis. To address this issue, we collect a new dataset of user reviews from multiple real-world scenarios such as e-retail, e-commerce, movie reviews, and social media. Due to the heterogeneity and complexity of this multi-scenario review data, we propose a self-distillation contrastive learning method. Specifically, we utilize two EMA (exponential moving average) models to generate soft labels as additional supervision. Additionally, we introduce the prototypical supervised contrastive learning module to reduce the variability of data in different scenarios by pulling in representations of the same class. Our method has proven to be extremely competitive, outperforming other advanced methods. Specifically, our method achieves an 87.44% F1 score, exceeding the performance of current advanced methods by 1.07%. Experimental results, including examples and visualization analysis, further demonstrate the superiority of our method. Full article

Image super-resolution (SR) technology can improve the resolution of images and provide clearer and more reliable remote sensing images of high quality to better serve the subsequent applications. However, when reconstructing high-frequency feature areas of remote sensing images, existing SR reconstruction methods are prone to artifacts that affect visual effects and make it difficult to generate real texture details. In order to address this issue, a texture-detail-enhanced dense generative adversarial network (TDEGAN) for remote sensing image SR is presented. The generator uses multi-level dense connections, residual connections, and Shuffle attention (SA) to improve the feature extraction ability. A PatchGAN-style discrimination network is designed to effectively perform local discrimination and helps the network generate rich, detailed features. To reduce the impact of artifacts, we introduce an artifact loss function, which is combined with the exponential moving average (EMA) technique to distinguish the artifacts generated from the actual texture details through local statistics, which can help the network reduce artifacts and generate more realistic texture details. Experiments show that TDEGAN can better restore the texture details of remote sensing images and achieves certain advantages in terms of evaluation indicators and visualization. Full article

In this article, we describe a performance comparison conducted between several digital filters intended to mitigate the intrinsic noise observed in magnetometers. The considered filters were used to smooth the control signals derived from the magnetometers, which were present in an autonomous forestry machine. Three moving average FIR filters, based on rectangular Bartlett and Hanning windows, and an exponential moving average IIR filter were selected and analyzed. The trade-off between the noise reduction factor and the latency of the proposed filters was also investigated, taking into account the crucial importance of latency on real-time applications and control algorithms. Thus, a maximum latency value was used in the filter design procedure instead of the usual filter order. The experimental results and simulations show that the linear decay moving average (LDMA) and the raised cosine moving average (RCMA) filters outperformed the simple moving average (SMA) and the exponential moving average (EMA) in terms of noise reduction, for a fixed latency value, allowing a more accurate heading angle calculation and position control mechanism for autonomous and unmanned ground vehicles (UGVs). Full article

Traditional fish farming methods suffer from backward production, low efficiency, low yield, and environmental pollution. As a result of thorough research using deep learning technology, the industrial aquaculture model has experienced gradual maturation. A variety of complex factors makes it difficult to extract effective features, which results in less-than-good model performance. This paper proposes a fish detection method that combines a triple attention mechanism with a You Only Look Once (TAM-YOLO)model. In order to enhance the speed of model training, the process of data encapsulation incorporates positive sample matching. An exponential moving average (EMA) is incorporated into the training process to make the model more robust, and coordinate attention (CA) and a convolutional block attention module are integrated into the YOLOv5s backbone to enhance the feature extraction of channels and spatial locations. The extracted feature maps are input to the PANet path aggregation network, and the underlying information is stacked with the feature maps. The method improves the detection accuracy of underwater blurred and distorted fish images. Experimental results show that the proposed TAM-YOLO model outperforms YOLOv3, YOLOv4, YOLOv5s, YOLOv5m, and SSD, with a mAP value of 95.88%, thus providing a new strategy for fish detection. Full article

In response to the high breakage rate of pigeon eggs and the significant labor costs associated with egg-producing pigeon farming, this study proposes an improved YOLOv8-PG (real versus fake pigeon egg detection) model based on YOLOv8n. Specifically, the Bottleneck in the C2f module of the YOLOv8n backbone network and neck network are replaced with Fasternet-EMA Block and Fasternet Block, respectively. The Fasternet Block is designed based on PConv (Partial Convolution) to reduce model parameter count and computational load efficiently. Furthermore, the incorporation of the EMA (Efficient Multi-scale Attention) mechanism helps mitigate interference from complex environments on pigeon-egg feature-extraction capabilities. Additionally, Dysample, an ultra-lightweight and effective upsampler, is introduced into the neck network to further enhance performance with lower computational overhead. Finally, the EXPMA (exponential moving average) concept is employed to optimize the SlideLoss and propose the EMASlideLoss classification loss function, addressing the issue of imbalanced data samples and enhancing the model’s robustness. The experimental results showed that the F1-score, mAP50-95, and mAP75 of YOLOv8-PG increased by 0.76%, 1.56%, and 4.45%, respectively, compared with the baseline YOLOv8n model. Moreover, the model’s parameter count and computational load are reduced by 24.69% and 22.89%, respectively. Compared to detection models such as Faster R-CNN, YOLOv5s, YOLOv7, and YOLOv8s, YOLOv8-PG exhibits superior performance. Additionally, the reduction in parameter count and computational load contributes to lowering the model deployment costs and facilitates its implementation on mobile robotic platforms. Full article

Electrodermal Activity (EDA) has shown great potential for emotion recognition and the early detection of physiological anomalies associated with stress. However, its non-stationary nature limits the capability of current analytical and detection techniques, which are highly dependent on signal stability and controlled environmental conditions. This paper proposes a framework for EDA normalization based on the exponential moving average (EMA) with outlier removal applicable to non-stationary heteroscedastic signals and a novel set of features for analysis. The normalized time series preserves the morphological and statistical properties after transformation. Meanwhile, the proposed features expand on typical time-domain EDA features and profit from the resulting normalized signal properties. Parameter selection and validation were performed using two different EDA databases on stress assessment, accomplishing trend preservation using windows between 5 and 20 s. The proposed normalization and feature extraction framework for EDA analysis showed promising results for the identification of noisy, relaxed and arousal-like patterns in data with conventional clustering approaches like K-means over the aforementioned normalized features. Full article