Search

Objective.Artificial neural networks (ANNs) are state-of-the-art tools for modeling and decoding neural activity, but deploying them in closed-loop experiments with tight timing constraints is challenging due to their limited support in existing real-time frameworks. Researchers need a platform that fully supports high-level languages for running ANNs (e.g. Python and Julia) while maintaining support for languages that are critical for low-latency data acquisition and processing (e.g. C and C++).Approach.To address these needs, we introduce the Backend for Realtime Asynchronous Neural Decoding (BRAND). BRAND comprises Linux processes, termednodes, which communicate with each other in agraphvia streams of data. Its asynchronous design allows for acquisition, control, and analysis to be executed in parallel on streams of data that may operate at different timescales. BRAND uses Redis, an in-memory database, to send data between nodes, which enables fast inter-process communication and supports 54 different programming languages. Thus, developers can easily deploy existing ANN models in BRAND with minimal implementation changes.Main results.In our tests, BRAND achieved

Applications of terahertz (THz) imaging technologies have advanced significantly in the disciplines of biology, medical diagnostics, and nondestructive testing in the past several decades. Significant progress has been made in THz biomedical imaging, allowing for the label-free diagnosis of malignant tumors. Terahertz frequencies, which lie between those of the microwave and infrared, are highly sensitive to water concentration and are significantly muted by water. Terahertz radiation does not cause ionization of biological tissues because of its low photon energy. Recently, terahertz spectra, including spectroscopic investigations of cancer, have been reported at an increasing rate due to the growing interest in their biological applications sparked by these unique features. To improve cancer diagnosis with terahertz imaging, an appropriate differentiation technique is required to increased blood supply and localized rise in tissue water content that commonly accompany the presence of malignancy. Terahertz imaging has been found to benefit from structural alterations in afflicted tissues. This study provides an overview of terahertz technology and briefly discusses the use of terahertz imaging techniques in the detection of skin cancer. Research into the promise and perils of terahertz imaging will also be discussed. [ABSTRACT FROM AUTHOR]

Interpersonal interaction is essential to romantic couples. Understanding how gender impacts an individual's brain activities during intimate interaction is crucial. The present study examined gender differences in oxyhemoglobin (oxy-Hb) changes during real-time drawing interactions between members of romantic couples using non-invasive functional near-infrared spectroscopy (fNIRS). We analyzed the oxy-Hb concentrations of romantic couples engaged in interactive (i.e., chase and escape) and non-interactive (i.e., individual) drawing sessions. Our findings indicated that males (vs. females) exhibited more pronounced oxy-Hb concentrations in Broca's area, motor area, sensorimotor cortex, and temporal lobe areas than women in an interactive drawing task, suggesting a heightened goal-oriented engagement in social interaction. Significant positive correlations were found between oxy-Hb volumes of the temporal area and the Quality of Relationship Index (QRI), underscoring the impact of interpersonal dynamics on brain function during interactive tasks. This study deepens the understanding of gender differences in neural mechanisms in social interaction tasks and provides important insights for intimacy research. [ABSTRACT FROM AUTHOR]

Background: Stereotactic arrhythmia radioablation (STAR) is a novel treatment approach for refractory ventricular tachycardia (VT). The risk of treatment‐induced toxicity and geographic miss can be reduced with online MRI‐guidance on an MR‐linac. However, most VT patients carry cardiac implantable electronic devices (CIED), which compromise MR images. Purpose: Robust MR‐linac imaging sequences are required for cardiac visualization and accurate motion monitoring in presence of a CIED during MRI‐guided STAR. We optimized two clinically available cine sequences for cardiorespiratory motion estimation in presence of a CIED on a 1.5 T MR‐linac. The image quality, motion estimation accuracy, and geometric fidelity using these cine sequences were evaluated. Methods: Clinically available 2D balanced steady‐state free precession (bSSFP, voxel size = 3.0 ×$\times$ 3.0 ×$\times$ 10 mm3, Tscan = 96 ms, bandwidth (BW) = 1884 Hz/px) and T1${\rm T}_{1}$‐spoiled gradient echo (T1${\rm T}_{1}$‐GRE, voxel size = 4.0 ×$ \times$ 4.0 ×$ \times$ 10 mm3, Tscan = 97 ms, BW = 500 Hz/px) sequences were adjusted for real‐time cardiac visualization and cardiorespiratory motion estimation on a 1.5 T Unity MR‐linac (Elekta AB, Stockholm, Sweden), while complying with safety guidelines for MRI in presence of CIEDs (specific absorption rate <$ <$ 2 W/kg and dBdt<$\frac{dB}{dt}<$ 80 mT/s). Cine acquisitions were performed in five healthy volunteers, with and without an implantable cardioverter– defibrillator (ICD) placed on the clavicle, and a VT patient. Generalized divergence‐curl (GDC) deformable image registration (DIR) was used for automated landmark motion estimation in the left ventricle (LV). Gaussian processes (GP), a machine‐learning technique, was trained using GDC landmarks and deployed for real‐time cardiorespiratory motion prediction. B0$B_{0}$‐mapping was performed to assess geometric image fidelity in the presence of CIEDs. Results: CIEDs introduced banding artifacts partially obscuring cardiac structures in bSSFP acquisitions. In contrast, the T1${\rm T}_{1}$‐GRE was more robust to CIED‐induced artifacts at the expense of a lower signal‐to‐noise ratio. In presence of an ICD, image‐based cardiorespiratory motion estimation was possible for 85% (100%) of the volunteers using the bSSFP (T1${\rm T}_{1}$‐GRE) sequence. The in‐plane 2D root‐mean‐squared deviation (RMSD) range between GDC‐derived landmarks and manual annotations using the bSSFP (T1‐GRE) sequence was 3.1–3.3 (3.3–4.1) mm without ICD and 4.6–4.6 (3.2–3.3) mm with ICD. Without ICD, the RMSD between the GP‐predictions and GDC‐derived landmarks ranged between 0.9 and 2.2 mm (1.3–3.0 mm) for the bSSFP (T1‐GRE) sequence. With ICD, the RMSD between the GP‐predictions and GDC‐derived landmarks ranged between 1.3 and 2.2 mm (1.2–3.2 mm) using the bSSFP (T1‐GRE) sequence resulting in an RMSD‐increase of 42%–143% (bSSFP) and −61%–142% (T1‐GRE). Lead‐induced spatial distortions ranged between −0.2 and 0.2 mm (−0.7–1.2 mm) using the bSSFP (T1${\rm T}_{1}$‐GRE) sequence. The 98th percentile range of the spatial distortions in the gross target volume of the patient was between 0.0 and 0.4 mm (0.0–1.8 mm) when using bSSFP (T1${\rm T}_{1}$‐GRE). Conclusions: Tailored bSSFP and T1${\rm T}_{1}$‐GRE sequences can facilitate real‐time cardiorespiratory estimation using GP trained with GDC‐derived landmarks in the majority of landmark locations in the LV despite the presence of CIEDs. The need for high temporal resolution noticeably reduced achievable spatial resolution of the cine MRIs. However, the effect of the CIED‐induced artifacts is device, patient and sequence dependent and requires specific assessment per case. [ABSTRACT FROM AUTHOR]

Detecting Broken Rotor Bars (BBFs) in induction motors is critical for ensuring their reliable operation. While conventional methods, such as Fast Fourier Transform analysis of stator current spectra, have been widely used for BBF detection, they suffer from limitations like spectral leakage and low-frequency resolution. The Welch Transform is known for effectively reducing spectral leakage and noise when analyzing finite data. This paper presented an innovative FPGA-based architecture for real-time implementation of Welch transform for BBFs in induction motor diagnostics accompanied by a novel right-band-based detection technique, and the architectures are explained in detail. We conducted experiments to verify the effectiveness of the proposed architectures, including applying BBF faults under varying loads and severity levels. The results demonstrated the efficiency of our proposed architectures, as it was found that resource consumption rates were meagre, and error indicators were obvious. The results were displayed in real-time through a user-friendly graphical interface, demonstrating the practical effectiveness of the FPGA-based solution. [ABSTRACT FROM AUTHOR]

Traditional methods of personality assessment, such as questionnaires and interviews, rely on self-reporting and subjective interpretation, which can be influenced by biases and social desirability. A handwriting-based approach offers an alternative method that provides non-verbal cues and unconscious expressions, supplementing traditional methods and potentially offering more objective insights, especially with an automated approach, into personality traits. Non-real-time methods take longer due to manual analysis, and are often subjective and prone to bias, while real-time analysis with a convolutional neural network (CNN) model provides instant results. Real-time tools would have no human intervention, are more convenient and easily accessible. Reducing human intervention through real-time analysis with a CNN model enhances the reliability, objectivity, scalability, and speed of the handwriting assessment process, providing a significant advantage over traditional methods. The proposed system focuses on the use of deep learning (DL) techniques to determine a person's personality by analyzing their handwriting. With the use of deep learning, human intervention is much less, thus the variance in the results is also much less which ultimately increases precision in the final predictions or results. Not only does the project predict personality traits through handwriting analysis, but also does that in a real-time environment. Steps such as preprocessing, feature extraction, and label classification are involved in the prediction process. CNN model has been used in the proposed system, for the final Personality Prediction. The CNN model for handwriting analysis must balance speed and accuracy through efficient architecture design and parameter optimization, minimizing computational complexity and inference time. To achieve this challenge two models (or versions) were built, the second having better data augmentation with respect to image sizing to be given to the CNN Network. Performance using the evaluation metrics was calculated with a testing efficiency of 0.71 in model 1 and 0.74 in model 2 and 82% accuracy was obtained on this real-time data. [ABSTRACT FROM AUTHOR]

The field of fabric defect detection has undergone a transformative journey marked by the evolution of object detection models. From traditional approaches to advanced deep learning architectures, these models have addressed crucial challenges in the textile industry. YOLOv7-tiny model stands out as a remarkable solution, demonstrating unprecedented performance in fabric defect detection. Its enhanced architecture addresses key industry issues, including high-resolution images, small defect sizes, and imbalanced datasets. Therefore, the aim of this paper is to incorporate the YOLOv7 model with improvements to detect woven fabric defects in real time. Augmenting the Enhanced Residual Convolutional Network (ERCN) with extra Convolutional, batch normalization and leaky rectified linear unit (CBL) layers enhances hierarchical feature extraction, while the two-concatenation technique adds complexity for richer representations. Reducing CBL layers in Efficient layer aggregation networks-downgrade (ELAN-D) streamlines and optimizes, emphasizing a balanced approach in the YOLOv7-tiny model for targeted objectives. The improved YOLOv7-tiny model excels in achieving a delicate balance between accuracy and efficiency, vital for practical applications in the textile sector. This model's accuracy, with a mAP of 84% at a 0.50 threshold and 40% at 0.50:0.95 showed exceptional in comparisons to other models. The model also boasts a high accuracy of 98% and operates at a commendable detection speed of 90 fps, meeting real-time demands in fabric production. Recognizing defects as small as 1 mm, the YOLOv7-tiny model emerges as a pivotal tool in automating fabric defect detection and optimizing textile quality management processes. [ABSTRACT FROM AUTHOR]

Industry requires defect detection to ensure the quality and safety of products. In resource-constrained devices, real-time speed, accuracy, and computational efficiency are the most critical requirements for defect detection. This paper presents a novel approach for real-time detection of surface defects on LPG cylinders, utilising an enhanced YOLOv5 architecture referred to as GLDD-YOLOv5. The architecture integrates ghost convolution and ECA blocks to improve feature extraction with less computational overhead in the network's backbone. It also modifies the P3–P4 head structure to increase detection speed. These changes enable the model to focus more effectively on small and medium-sized defects. Based on comparative analysis with other YOLO models, the proposed method demonstrates superior performance. Compared to the base YOLOv5s model, the proposed method achieved a 4.6% increase in average accuracy, a 44% reduction in computational cost, a 45% decrease in parameter counts, and a 26% reduction in file size. In experimental evaluations on the RTX2080Ti, the model achieved an inference rate of 163.9 FPS with a total carbon footprint of 0.549 × 10−3 gCO2e. The proposed technique offers an efficient and robust defect detection model with an eco-friendly solution compatible with edge computing devices. [ABSTRACT FROM AUTHOR]

This study addresses the frequent convergence issues of satellite clocks within regional network, with a particular focus on the multifrequency advantages using data from 25 uniformly distributed reference stations across China. Experimental results demonstrate that incorporating the third frequency significantly enhances the accuracy of BDS-3 clock solutions, reducing the root mean square (RMS) by 44.5%. Additionally, employing a 2-min smoothing interval, multifrequency inclusion increases the wide-lane (WL) fixing rate by 30.4% at low elevation angles, which in turn leads to a marked improvement in narrow-lane (NL) ambiguity resolution. By leveraging phase-wide-lane observations, the stable wide-lane phase bias enables the continuous generation of inter-frequency clock bias (IFCB), ensuring reliable cyclic sequence construction even when satellites exit the observed region. The effectiveness of regional observable specific bias (OSB) on ambiguity resolution at the user level is highlighted, and over 95% of GPS, BDS-3, and Galileo NL fractional biases below 0.15 cycles could be achieved. Furthermore, the single-epoch convergence rates of multi-constellation precise point positioning (PPP) reach horizontal 91.9% and vertical 84.5% for multifrequency, a substantial improvement over the dual-frequency, which does not exceed 25%. [ABSTRACT FROM AUTHOR]

Sixth-generation (6G) networks are anticipated to achieve transformative advancements, characterized by extreme connectivity, deep integration with artificial intelligence (AI) and sensing, and airground integration. The evolution of 6G exhibits two major trends: ubiquitous intelligence and sustainability. The former aims to embed state-of-the-art AI technology into the 6G network, from the physical layers to applications, while the latter emphasizes reducing energy consumption while enhancing network performance to address environmental concerns. Despite the amazing progress in recent years, AI advancements come with substantial increases in data and computational overhead, posing critical challenges for integrating AI into sustainable 6G networks. First, high energy consumption from large datasets and heavyweight AI models contradicts 6G’s green goals. Second, the precise collection of large datasets, message delivery latency, and inference delays in AI models pose challenges for real-time tasks in 6G. Third, the uninterpretability and unpredictability of AI models complicate meeting the stringent requirements for controllable transmission in dynamic wireless environments. Addressing these challenges and achieving sustainable 6G with ubiquitous intelligence calls for a revolutionary design of 6G architecture and AI frameworks. To this end, this paper introduces a novel and practical methodology for green, real-time, and controllable 6G native intelligence, starting with knowledge graph (KG) analysis to extract small but critical datasets, followed by the development of distributed lightweight AI models, and the use of digital twins (DTs) to create precise replicas of physical 6G networks. This leads to a pervasive multi-level (PML)-AI framework supported by a task-centric, three-layer 6G architecture. The AI framework operates through non-real-time and real-time cycles, leveraging three key technologies: wireless data KGs for efficient data management, lightweight AI models for sub-millisecond real-time responsiveness, and DTs for AI pre-validation. A prototype system is built on the proposed 6G architecture and PML-AI framework, and experimental results show that data overhead is significantly reduced and real-time intelligence at the millisecond level can be realized. [ABSTRACT FROM AUTHOR]

Currently, the susceptibility of Global Navigation Satellite System (GNSS) signals underscores the importance of accurate GNSS time spoofing detection as a critical research area. Traditional spoofing detection methods have limitations in applicability, while the current learning-based algorithms are only applicable to the judgment of collected data, which is difficult to apply to real-time detection. In this paper, a real-time spoofing detection framework based on feature processing is proposed. The approach involves feature integration and correlation coefficient screening on each epoch of multi-satellite data. Additionally, special standardization strategy is employed to enhance the feasibility of real-time application. In the experimental phase, apart from utilizing the open dataset, an experimental platform is developed to generate dual-system data for experimentation purposes. Compared with the traditional clock difference detection method, this algorithm improves the detection performance by about 25%. Furthermore, the framework proposed can improve the detection F1 score of basic machine learning models and greatly reduce the computation time by more than ten times. On most datasets, models incorporating the framework achieved F1 scores of more than 99% and average response times of less than 10 μs. In summary, this study provides an effective intelligent solution for the application of real-time receiver spoofing detection. [ABSTRACT FROM AUTHOR]

This study offers a fresh technique for translating subtitles in sports events, addressing the issues of real-time translation with improved accuracy and efficiency. Different from standard methods, which often result in delayed or inaccurate subtitles, the proposed method integrates advanced annotation techniques and machine learning algorithms to increase subtitle recognition and extraction. Annotation techniques in this study include systematically labeling spoken elements like commentary and dialogue, enabling accurate subtitle recognition and real-time adjustments in live sports broadcasts to ensure both accuracy and contextual relevance. These novel ideas allow for seamless adjustments to multiple language types, including the voices of commentators, off-site hosts, and athletes, while maintaining critical information within strict word count limits. Key improvements include faster processing times and increased translation precision, which are crucial for the dynamic environment of live sports broadcasts. The study builds on past studies in audiovisual translation, specifically tailoring its strategy to the unique demands of sports media. By emphasizing the importance of clear and contextually appropriate real-time subtitles, this research presents significant advancements over existing methods, providing valuable insights for future translation projects in sports and similar contexts. The results contribute to a more effective subtitle translation framework, enhancing the accessibility and viewing experience for audiences during live sports events. [ABSTRACT FROM AUTHOR]

Automatic extraction of coal flow region of coal mine belt conveyor plays an important role in coal flow monitoring, and real-time control of belt speed through real-time accurate monitoring of coal flow, which realizes the purpose of energy saving and consumption reduction of belt conveyor. In this paper, a real-time semantic segmentation network with detail enhancement for pixel-level coal flow monitoring, called DENet, is proposed. First, to ensure the strong real-time performance of the network, a two-branch coding structure is used to extract the semantic information and spatial detail information. Second, to improve the feature representation of spatial detail information, we design the Parameter-free Attention-Guided Enhancement Module (PF-AGEM) and the detail enhancement module (DEM), which fully integrate the semantic information features in the semantic branch into the detail branch and further enhance the detail features. Third, we design the multi-scale channel attention (MSCA) module in the semantic branch to extract the semantic information features of small targets earlier in the high-resolution feature maps, which solves the problem that the semantic information features of small targets are easily lost in the low-resolution feature maps. Finally, we propose a selective feature fusion module (FFM) to better realize the fusion of semantic information and spatial detail information. Experimental results show that the proposed DENet achieves a mean intersection over union (mIoU) of 96.23% at 87.1 frames per second (FPS) on the Coal Flow Segmentation (CFS) dataset and 74.9% mIoU at 207 FPS on the Camvid dataset, which is competitive with the state-of-the-art real-time semantic segmentation models. [ABSTRACT FROM AUTHOR]

The exponential growth of Internet of Things (IoT) devices underscores the need for robust security measures against cyber-attacks. Extensive research in the IoT security community has centered on effective traffic detection models, with a particular focus on anomaly intrusion detection systems (AIDS). This paper specifically addresses the preprocessing stage for IoT datasets and feature selection approaches to reduce the complexity of the data. The goal is to develop an efficient AIDS that strikes a balance between high accuracy and low detection time. To achieve this goal, we propose a hybrid feature selection approach that combines filter and wrapper methods. This approach is integrated into a two-level anomaly intrusion detection system. At level 1, our approach classifies network packets into normal or attack, with level 2 further classifying the attack to determine its specific category. One critical aspect we consider is the imbalance in these datasets, which is addressed using the Synthetic Minority Over-sampling Technique (SMOTE). To evaluate how the selected features affect the performance of the machine learning model across different algorithms, namely Decision Tree, Random Forest, Gaussian Naive Bayes, and k-Nearest Neighbor, we employ benchmark datasets: BoT-IoT, TON-IoT, and CIC-DDoS2019. Evaluation metrics encompass detection accuracy, precision, recall, and F1-score. Results indicate that the decision tree achieves high detection accuracy, ranging between 99.82 and 100%, with short detection times ranging between 0.02 and 0.15 s, outperforming existing AIDS architectures for IoT networks and establishing its superiority in achieving both accuracy and efficient detection times. [ABSTRACT FROM AUTHOR]

Copyright of Machine Tool & Hydraulics is the property of Guangzhou Mechanical Engineering Research Institute (GMERI) and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Lung ultrasound is an increasingly utilized non-invasive imaging modality for assessing lung condition but interpreting it can be challenging and depends on the operator's experience. To address these challenges, this work proposes an approach that combines artificial intelligence (AI) with feature-based signal processing algorithms. We introduce a specialized deep learning model designed and trained to facilitate the analysis and interpretation of lung ultrasound images by automating the detection and location of pulmonary features, including the pleura, A-lines, B-lines, and consolidations. Employing Convolutional Neural Networks (CNNs) trained on a semi-automatically annotated dataset, the model delineates these pulmonary patterns with the objective of enhancing diagnostic precision. Real-time post-processing algorithms further refine prediction accuracy by reducing false-positives and false-negatives, augmenting interpretational clarity and obtaining a final processing rate of up to 20 frames per second with accuracy levels of 89% for consolidation, 92% for B-lines, 66% for A-lines, and 92% for detecting normal lungs compared with an expert opinion. [ABSTRACT FROM AUTHOR]

Summary: Optimizing energy consumption and maximizing throughput in multi‐core real‐time architectures through dynamic task scheduling is a critical design challenge. While significant attention has been devoted to addressing this challenge in the domain of real‐time multi‐core scheduling, the focus has primarily centered on considering periodic tasks as independent. However, the existing literature notably lacks comprehensive study of scheduling methodologies on multi‐core systems that consider dependent tasks, though typical real‐time systems execute tasks that share resources. Earlier studies have predominantly examined scenarios involving random new tasks and task instances (jobs), which are executed in different power levels. Each task (and job) has distinct execution time corresponding to each power level. By considering these parameters (power levels and execution times of jobs), various combinations of energy signatures have been found to attain an optimum system state. Building upon this prior research, our paper extends the scope to encompass task scheduling in multi‐core systems with task dependencies. We introduce a novel approach that categorizes dependent tasks into ASAP (as soon as possible) and ALAP (as late as possible) groups, prioritizing task execution based on task mobility—defined as the disparity between the last cycle the task can be scheduled in and the current cycle. Furthermore, our model demonstrates an approach for efficient scheduling of sporadic and aperiodic tasks within this framework. Through experimental validation using randomized task sets, our results indicate that the proposed model achieves a minimum of 5% reduction in normalized total energy consumption compared to existing methodologies. [ABSTRACT FROM AUTHOR]

Background: Blood cultures are essential for diagnosing bloodstream infections, but current phenotypic tests for antimicrobial resistance (AMR) provide limited information. Oxford Nanopore Technologies introduces nanopore sequencing with adaptive sampling, capable of real-time host genome depletion, yet its application directly from blood cultures remains unexplored. This study aimed to identify pathogens and predict AMR using nanopore sequencing. Methods: In this cross-sectional genomic study, 458 positive blood cultures from bloodstream infection patients in central Taiwan were analyzed. Parallel experiments involved routine microbiologic tests and nanopore sequencing with a 15-h run. A bioinformatic pipeline was proposed to analyze the real-time sequencing reads. Subsequently, a comparative analysis was performed to evaluate the performance of species identification and AMR prediction. Results: The pipeline identified 76 species, with 88 Escherichia coli, 74 Klebsiella pneumoniae, 43 Staphylococcus aureus, and 9 Candida samples. Novel species were also discovered. Notably, precise species identification was achieved not only for monomicrobial infections but also for polymicrobial infections, which was detected in 23 samples and further confirmed by full-length 16S rRNA amplicon sequencing. Using a modified ResFinder database, AMR predictions showed a categorical agreement rate exceeding 90% (3799/4195) for monomicrobial infections, with minimal very major errors observed for K. pneumoniae (2/186, 1.1%) and S. aureus (1/90, 1.1%). Conclusions: Nanopore sequencing with adaptive sampling can directly analyze positive blood cultures, facilitating pathogen detection, AMR prediction, and outbreak investigation. Integrating nanopore sequencing into clinical practices signifies a revolutionary advancement in managing bloodstream infections, offering an effective antimicrobial stewardship strategy, and improving patient outcomes. [ABSTRACT FROM AUTHOR]

Normal people's interaction and communication are easier than those with disabilities such as hearing and speech, which are very complicated; hence, the use of sign language plays a crucial role in bridging this gap in communication. While previous attempts have been made to solve this problem using deep learning techniques, including convolutional neural networks (CNNs), support vector machine (SVM), and K-nearest neighbours (KNN), these have low accuracy or may not be employed in real time. This work addresses both issues: improving upon prior limitations and extending the challenge of classifying characters in Indian sign language (ISL). Our system, which can recognize 23 hand gestures of ISL through a purely camera-based approach, eliminates expensive hardware like hand gloves, thus making it economical. The system yields an accuracy of 97.5% on the training dataset, utilizing a pre-trained VGG16 CNN optimized by the Adam optimizer and cross-entropy loss function. These results clearly show how effective transfer learning is in classifying ISL and its possible real-world applications. [ABSTRACT FROM AUTHOR]

There are currently many in-field methods for estimating soil properties (e.g., pH, texture, total C, total N) available in precision agriculture, but each have their own level of suitability and only a few can be used for direct determination of plant-available nutrients. As promising approaches for reliable in-field use, this review provides an overview of electromagnetic, conductivity-based, and electrochemical techniques for estimating plant-available soil nutrients and pH. Soil spectroscopy, conductivity, and ion-specific electrodes have received the most attention in proximal soil sensing as basic tools for precision agriculture during the last two decades. Spectral soil sensors provide indication of plant-available nutrients and pH, and electrochemical sensors provide highly accurate nitrate and pH measurements. This is currently the best way to accurately measure plant-available phosphorus and potassium, followed by spectral analysis. For economic and practicability reasons, the combination of multi-sensor in-field methods and soil data fusion has proven highly successful for assessing the status of plant-available nutrients in soil for precision agriculture. Simultaneous operation of sensors can cause problems for example because of mutual influences of different signals (electrical or mechanical). Data management systems provide relatively fast availability of information for evaluation of soil properties and their distribution in the field. For rapid and broad adoption of in-field soil analyses in farming practice, in addition to accuracy of fertilizer recommendations, certification as an official soil analysis method is indispensable. This would strongly increase acceptance of this innovative technology by farmers. [ABSTRACT FROM AUTHOR]

BDS-3 geostationary orbit (GEO) satellites broadcast PPP-B2b real-time precise products to support real-time precise point positioning (RT-PPP) without dependence on the ground network communication. Since all BDS-3 satellites can provide multi-frequency signals, we investigate the effect of different multi-frequency integrations and the refinement of multi-frequency stochastic model for PPP-B2b RT-PPP. Given that the retrieval of PPP-B2b real-time precise corrections relies on the B1C and B2b frequencies, this study first investigates the kinematic uncombined (UC) RT-PPP performance of the B1C/B2b integration with a comparison to the conventional B1I/B3I integration. The results indicate that the convergence time (with a convergence threshold of 20 cm) of the RT-PPP with the B1C/B2b integration is shortened by 16 %, 15 %, and 12 % over the B1I/B3I integration in the east, north, and up directions, respectively. Further, this study compares the kinematic UC RT-PPP performance of the B1C/B2b dual-frequency integration, B1C/B2b/B3I triple-frequency integration, and B1C/B2b/B3I/B1I/B2a five-frequency integration. Compared with the dual-frequency integration, the positioning performance of the triple-frequency integration is slightly improved. By contrast, the performance improvement of the five-frequency integration is more significant, which can reach 6 %, 32 %, and 9 % for the convergence time, and 11 %, 9 %, and 6 % for the positioning accuracy in three directions, respectively. To further enhance the multi-frequency PPP performance, a refined stochastic model for the BDS-3 five-frequency integrated PPP-B2b RT-PPP is proposed by taking the inconsistency of signal quality among multi-frequency signals into account. After applying the refined stochastic model, the convergence time of the five-frequency kinematic RT-PPP is shortened by 10 %, 14 %, and 9 % to 32, 7, and 19 min in east, north, and up directions, respectively. The positioning accuracy (the root mean square of all the positioning errors excluding the first two hours) of the five-frequency RT-PPP with the refined stochastic model can reach the optimal level, which is 9.9, 6.5, and 14.0 cm in three directions, respectively. [ABSTRACT FROM AUTHOR]

Identifying traffic signs is an essential part of traffic safety and self-driving systems. In real life, the driving environment is changing, making detecting traffic signs wisely and economically vital. The traffic sign detection problem has several small objects and complex ambient interference. The detecting situation also requires a practical and lightweight detection model. This study proposes a new lightweight model, the enhanced Ghost-YOLOv8, based on lightweight modules GhostConv and C3Ghost, based on the YOLOv8 model. It used a light method to extract the features, significantly speeding up inference. In addition to small, medium, and large objects, the head was expanded to include a new multi-scale feature extraction module layer focused on x-small. The experiment results show that when using the German Traffic Sign Detection Benchmark (GTSDB) dataset with three classes, the enhanced Ghost-YOLOv8 has mAP (0.50) of 99.4%and has fewer computations than the YOLOv8 model by 155.2 GFLOPs and has 18.6 Mparameters, which represents only 27.3% from the parameters used in the base model. Also, we suggested a new dataset called the GTSDB-43 dataset, which expanded the number of classes on the GTSDB dataset from three or four main classes to 43 classes and mentioned their main category type simultaneously. Compared with notable algorithms, this method's accuracy and speed are competitive. [ABSTRACT FROM AUTHOR]

In the application of image super-resolution (SR) based on field-programmable gate array (FPGA), depthwise separable convolution is widely utilized. However, existing network designs overly simplify the structures used for deep feature extraction to conserve on-chip memory resources, which compromises network performance. Despite this, they still consume relatively large amounts of hardware resources. This study adopts a hardware-software co-design approach, proposing a hardware-friendly algorithm based on recurrent convolution neural network (RCNN), and implemented on FPGA. The algorithm employs a lightweight information multi-distillation block (LIMDB) to deeply extract and distill feature information, enhancing the network’s receptive field and improving its capability to extract feature information. Concurrently, the network reduces the dimensions of hidden state, enhancing performance and decreasing data transmission. Experimental results demonstrate that our lightweight deep recurrent convolution network (LDRCN) significantly outperforms other methods on common datasets. At the hardware level, we designed an efficient pipelining structure that combines skip connection and line buffer shared storage with a lossless segmentation computational strategy to reduce on-chip memory usage. Additionally, a method for finding mixed-precision quantization bit widths was designed to significantly reduce computational resource consumption while ensuring accuracy. The synthesis results on the ZYNQ XCZU9EG platform show that, compared to similar RNN-based works, the computation speed increased by 24.3% and DSP usage decreased by 43.5%, achieving ultra-high-definition (UHD) video output at 4K 95 fps. [ABSTRACT FROM AUTHOR]

This research focuses on the application of Nginx Server to monitor the ripeness level of tea leaves in real-time using RTMP module and IP camera. In the tea industry, monitoring the ripeness of tea leaves is essential to ensure the quality of the final product. To achieve this goal, a system was built using Nginx as the main server that manages the real-time video transmission of IP cameras installed in tea plantations. The RTMP module is used to transmit streaming video efficiently and reliably. The implementation of this system involves several stages, including the installation and configuration of the Nginx server, the integration of the RTMP module, and the installation of the IP cameras in strategic locations. The video data collected was analyzed to determine the maturity level of tea leaves based on certain visual indicators. The results of this study show that the use of Nginx server with RTMP module can provide an effective and efficient solution for realtime monitoring of tea leaf ripeness, which can help farmers and tea producers in improving their quality and productivity. These findings make important contributions in the field of smart agriculture and IoT-based monitoring technologies, and open up opportunities for the development of similar systems in other agricultural contexts. [ABSTRACT FROM AUTHOR]

Background: To date, most intradialytic hypotension (IDH) studies have proposed technologies to comprehensively predict the occurrence of IDH using the patient's baseline information and ultrafiltration (UF) information, but it is difficult to apply the technologies while identifying the patient's condition in real time. Methods: In this study, we propose an IDH indicator that uses heart rate (HR) change information to identify the patient's condition in real time and visually shows whether IDH has occurred. The data used were collected from 40 dialysis patients in a clinical trial conducted in the Artificial Kidney Unit at Yeungnam University Medical Center, Korea, from 18 July to 29 November 2023. Results: The IDH indicator infers changes in the patient's blood pressure during dialysis by analyzing the upper and lower maximum HRs based on the real-time average HR. Medical staff can respond to IDH in real time by looking at the IDH indicator, which visually expresses changes in the patient's HR. In addition, we propose a multilayer perceptron structure that inputs upper and lower maximum HR information based on the average HR for the time interval accumulated in real time. In learning using 40 min of data up to 5 min before IDH occurred, models using two and five layers showed excellent performance, with accuracy of 88.6% and 85.2%, respectively. Conclusions: By combining IDH visual indicators and the multi-layer perceptron method, medical staff can effectively respond to IDH in real time. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Harbin Institute of Technology. Social Sciences Edition / Haerbin Gongye Daxue Xuebao. Shehui Kexue Ban is the property of Harbin Institute of Technology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

The rapid proliferation of the Internet of Things (IoT) has significantly increased the risk of Distributed Denial of Service (DDoS) attacks, threatening the reliability, availability, and security of services and infrastructure. To address these challenges, this study introduces a novel, integrated framework combining fog computing, machine learning, and lightweight encryption to enhance offline training and real-time detection of DDoS attacks in IoT environments. Our approach differs from existing methods by leveraging an offline phase for model training on recent DDoS patterns. This enables accurate, scalable detection when the model is deployed in the online fog layer. This two-phase strategy ensures timely and resource-efficient threat mitigation. In the offline phase, we extract four critical packet features (Src_IP, Port_IP, Dst_IP, Dst_Port) from the CICDDoS2019 and Edge-IIoTset datasets. We then apply Chi-Square and entropy-based feature analysis, followed by synthetic minority oversampling (SMOTE), to address class imbalance. Three core classifiers--Random Forest (RF), Support Vector Machine (SVM), and Decision Tree (DT)--are trained to detect a variety of DDoS attacks (SYN, UDP, HTTP, and TCP) with high accuracy. The online phase deploys the trained model at the fog layer, employing the Speck lightweight encryption algorithm and Elliptic-Curve Diffie-Hellman (ECDH) for secure end-to-end communication. A voting mechanism among classifiers enhances detection reliability. The experiments proved that the framework achieves almost perfect detection accuracy (100% in most scenarios), surpassing current methods in accuracy, scalability, and applicability in resource-limited IoT environments. In addition, network performance metrics (throughput, latency, execution time, response time) confirm the solution's efficiency. This research provides a secure, adaptive, high-performance DDoS attack detection system for IoT systems, laying the foundation for future studies to expand attack coverage, improve real-time performance, and investigate more robust encryption methods. [ABSTRACT FROM AUTHOR]

Multicore systems introduced a performance increase over previous monocore systems. As these systems are increasingly finding application in critical domains, it arises a necessity to develop novel methodologies for their efficient resource allocation. In addition, it is particularly important to consider the criticality of applications when scheduling such systems. In multicore systems, scheduling also includes the allocation of tasks to cores. In architectures based on spatial and temporal partitioning, it is also necessary to allocate partitions. Consideration of all these variables when scheduling a critical multicore partitioned system is a major challenge. In this article, a hypervisor partitioned framework for mixed-criticality systems is proposed. In this sense, the allocation process has been split in two different parts. The initial phase will allocate tasks to partitions according to the criticality of the system. This is achieved through the implementation of a Mixed-Integer Linear Programming (MILP) algorithm. The second phase involves the allocation of tasks to cores, employing both, an additional MILP algorithm and a modified worst fit decrease utilisation approach. Experimental results show that the combination of both strategies leads to feasible scheduling and, in addition, to a reduction of the overhead introduced by the hypervisor. [ABSTRACT FROM AUTHOR]

Humanoid robots are designed to mimic human structures and utilize cameras to process visual input to identify surrounding objects. However, previous studies have focused solely on object detection, overlooking both the complexities of real-world implementation and the significance of calculating the distance between objects and the robot. This study proposes a system that employs the you only look once (YOLO) algorithm to detect various objects in the proximity of a robot. Using a dataset of primary data collected in a laboratory, the detected objects are from 12 classes, including humans, chairs, tables, cabinets, computers, books, doors, bottles, eggs, learning modules, cups, and hands, with each class comprising 1500 data points. Two YOLO architectures, namely tiny YOLOv3 and tiny YOLOv4, are assessed for their performance in object detection, with the tiny YOLOv4 demonstrating a superior accuracy of 82.99% compared to tiny YOLOv3. Evaluation under simulated conditions yields an accuracy of 74.16%, while in real-time scenarios, accuracies are 61.66% under bright conditions and 38.33% under dim conditions, affirming tiny YOLOv4's efficacy. Moreover, this study reveals an average error distance of 31% between an object and the robot in real-time conditions. The developed system enhances human-robot interaction capabilities via data transmission. [ABSTRACT FROM AUTHOR]

Wireless Sensor Networks (WSNs) are expressively utilized in various real-time control and monitoring applications. WSNs have been expanded considering the necessities in industrial time-bounded applications to support the dependable and time-bound delivery of data. Recently, Machine Learning (ML) algorithms have been used to address various WSN-related issues. The use of ML techniques supports dynamically modifying MAC settings based on traffic patterns and network conditions. In WSNs to control the communication between a large numbers of tiny, low-power sensor nodes while preserving energy and reducing latency, effective MAC protocols are essential. This paper addresses the ML centered prioritybased self-organized MAC (ML-MAC) protocol to provide a priority-based transmission system to ensure the timely delivery of critical data packets. In this research, depending upon the predictions of the ML model, the MAC parameters are dynamically adjusted to find priority-based channel access and the optimal routing path to meet the deadline of critical data packets. From the result analysis, the average throughput and delay of the proposed ML-MAC algorithm outperforms the existing I-MAC protocol. [ABSTRACT FROM AUTHOR]

The monitoring of acetone in exhaled breath is expected to provide a noninvasive and painless method for dynamic monitoring of summarized physiological metabolic status during obesity treatment. Although the commonly used Mass Spectrometry (MS) technology has high accuracy, the long detection time and large equipment size limit the application of daily bedside detection. As for the real-time and accurate detection of acetone, the gas sensor has become the best choice of gas detection technology, but it is easy to be disturbed by water vapor in breath gas. An integrated breath gas detection system based on cavity ring-down spectroscopy (CRDS) is reported in this paper, which is a laser absorption spectroscopy technique with high-sensitivity detection and absolute quantitative analysis. The system uses a 266 nm single-wavelength ultraviolet laser combined with a breath gas pretreatment unit to effectively remove the influence of water vapor. The ring-down time of this system was 1.068 μs, the detection sensitivity was 1 ppb, and the stability of the system was 0.13%. The detection principle of the integrated breath gas detection system follows Lambert–Beer's law, which is an absolute measurement with very high detection accuracy, and was further validated by Gas Chromatography–Mass Spectrometer (GC-MS) testing. Significant differences in the response of the integrated breath gas detection system to simulated gases containing different concentrations of acetone indicate the potential of the system for the detection of trace amounts of acetone. Meanwhile, the monitoring of acetone during obesity treatment also signifies the feasibility of this system in the dynamic monitoring of physiological indicators, which is not only important for the optimization of the obesity treatment process but also promises to shed further light on the interaction between obesity treatment and physiological metabolism in medicine. [ABSTRACT FROM AUTHOR]

In this generation, deep learning techniques are widely used for sign language prediction. In this paper, a deep learning model is proposed for American sign language detection using webcam images and transfer learning. The particular model is designed for a real-time sign language detection. The author claims 98% of accuracy for this designed model, when trained with a total of 15 images for each gesture. Jupyter notebook is used as the environment for working out this research. Cuda, cudnn graphic processor upgraders are also utilised in this research for training the model. To make real-time detections easy, a local environment is used rather than a cloud system for implementing the code. The main aim of this research work is to create a model in order to identify and detect the sign language alphabets and some very frequently used gestures. The model is designed based on deep learning by using the convolutional neural networks and single shot detector algorithm to surpass the difficulty that is faced by the speech impaired and normal people. [ABSTRACT FROM AUTHOR]

We present a hard-real-time multi-resolution mesh shape deformation technique for skeleton-driven soft-body characters. Producing mesh deformations at multiple levels of detail is very important in many applications in computer graphics. Our work targets applications where the multi-resolution shapes must be generated at fast speeds ("hard-real-time", e.g., a few milliseconds at most and preferably under 1 millisecond), as commonly needed in computer games, virtual reality and Metaverse applications. We assume that the character mesh is driven by a skeleton, and that high-quality character shapes are available in a set of training poses originating from a high-quality (slow) rig such as volumetric FEM simulation. Our method combines multi-resolution analysis, mesh partition of unity, and neural networks, to learn the pre-skinning shape deformations in an arbitrary character pose. Combined with linear blend skinning, this makes it possible to reconstruct the training shapes, as well as interpolate and extrapolate them. Crucially, we simultaneously achieve this at hard real-time rates and at multiple mesh resolution levels. Our technique makes it possible to trade deformation quality for memory and computation speed, to accommodate the strict requirements of modern real-time systems. Furthermore, we propose memory layout and code improvements to boost computation speeds. Previous methods for realtime approximations of quality shape deformations did not focus on hard real-time, or did not investigate the multi-resolution aspect of the problem. Compared to a "naive" approach of separately processing each hierarchical level of detail, our method offers a substantial memory reduction as well as computational speedups. It also makes it possible to construct the shape progressively level by level and interrupt the computation at any time, enabling graceful degradation of the deformation detail. We demonstrate our technique on several examples, including a stylized human character, human hands, and an inverse-kinematics-driven quadruped animal. [ABSTRACT FROM AUTHOR]