Hardware Configuration Research Articles

Land Classification Plugin for QGIS Using Pix2Pix

Abstract. Land cover classification is critical in various fields, including environmental monitoring, urban planning, and ecological assessment, facilitating informed decision-making processes. Traditional land cover classification methods often involve labor-intensive and time-consuming processes, relying on manual intervention and predefined algorithms. The emergence of deep learning techniques, particularly convolutional neural networks (CNNs), offers a promising solution to automate this process, albeit with complexities in implementation. This study addresses the limitations of existing Geographic Information System (GIS) software and plugins by proposing a novel approach utilizing the Pix2Pix architecture, a type of CNN, for automated land cover classification. The proposed Land Classification Plugin (LCP) integrates seamlessly with QGIS, offering an end-to-end solution for generating classified static maps. The methodology involves preprocessing data, utilizing the Pix2Pix model for image segmentation, and post-processing to produce georeferenced outputs. The development of the LCP involved extensive software and hardware configurations, including essential components like GDAL/OGR, PyTorch, and OpenCV. The plugin's architecture comprises a user-friendly interface for region selection, clipping, and classification aided by the Pix2Pix model. A layout manager feature also allows for the creation of composite maps for enhanced visualization. The accuracy assessment of the LCP demonstrated an overall accuracy of 83.40% across diverse land cover classes, indicating its efficacy in classification tasks. The plugin's capabilities offer significant potential for applications in land management, environmental surveillance, and urban planning, revolutionizing current practices in land cover classification within the realm of GIS software.

An in-air synthetic aperture sonar dataset of target scattering in environments of varying complexity

This paper describes a synthetic aperture sonar (SAS) dataset collected in-air consisting of four types of targets in four environments of different complexity. The in-air laboratory based experiments produced data with a level of fidelity and ground truth accuracy that is not easily attainable in data collected underwater. The range of complexity, high level of data fidelity, and accurate ground truth provides a rich dataset with acoustic features on multiple scales. It can be used to develop new signal-processing and image reconstruction algorithms, as well as machine learning models for object detection and classification. It may also find application in model verification and validation for acoustic simulators. The dataset consists of raw acoustic time series returns, associated environmental conditions, hardware configuration, array motion, as well as the reconstructed imagery.

Edge Computing for AI-Based Brain MRI Applications: A Critical Evaluation of Real-Time Classification and Segmentation.

Medical imaging plays a pivotal role in diagnostic medicine with technologies like Magnetic Resonance Imagining (MRI), Computed Tomography (CT), Positron Emission Tomography (PET), and ultrasound scans being widely used to assist radiologists and medical experts in reaching concrete diagnosis. Given the recent massive uplift in the storage and processing capabilities of computers, and the publicly available big data, Artificial Intelligence (AI) has also started contributing to improving diagnostic radiology. Edge computing devices and handheld gadgets can serve as useful tools to process medical data in remote areas with limited network and computational resources. In this research, the capabilities of multiple platforms are evaluated for the real-time deployment of diagnostic tools. MRI classification and segmentation applications developed in previous studies are used for testing the performance using different hardware and software configurations. Cost-benefit analysis is carried out using a workstation with a NVIDIA Graphics Processing Unit (GPU), Jetson Xavier NX, Raspberry Pi 4B, and Android phone, using MATLAB, Python, and Android Studio. The mean computational times for the classification app on the PC, Jetson Xavier NX, and Raspberry Pi are 1.2074, 3.7627, and 3.4747 s, respectively. On the low-cost Android phone, this time is observed to be 0.1068 s using the Dynamic Range Quantized TFLite version of the baseline model, with slight degradation in accuracy. For the segmentation app, the times are 1.8241, 5.2641, 6.2162, and 3.2023 s, respectively, when using JPEG inputs. The Jetson Xavier NX and Android phone stand out as the best platforms due to their compact size, fast inference times, and affordability.

A Review of Factors Affecting Radiation Dose and Image Quality in Coronary CTA Performed with Wide-Detector CT

Compared with traditional invasive coronary angiography (ICA), coronary CT angiography (CCTA) has the advantages of being rapid, economical, and minimally invasive. The wide-detector CT, with its superior temporal resolution and robust three-dimensional reconstruction technology, thus enables CCTA in patients with high heart rates and arrhythmias, leading to a high potential for clinical application. This paper systematically summarizes wide-detector CT hardware configurations of various vendors routinely used for CCTA examinations and reviews the effects of patient heart rate and heart rate variability, scanning modality, reconstruction algorithms, tube voltage, and scanning field of view on image quality and radiation dose. In addition, novel technologies in the field of CT applied to CCTA examinations are also presented. Since this examination has a diagnostic accuracy that is highly consistent with ICA, it can be further used as a routine examination tool for coronary artery disease in clinical practice.

RECOMMENDATIONS FOR THE DEPLOYMENT OF EQUIPMENT FOR TRAINING NEURAL NETWORKS

The paper presents recommendations for choosing a hardware resource when training neural networks. Analytical expressions are considered for calculating the computational complexity when deploying direct propagation neural networks of various dimensions. Some problems arising in the selection and configuration of hardware on which neural networks are trained are formulated. An assessment of the use of libraries/frameworks of various neural networks is given. A set of recommendations for the mutual arrangement of hardware modules is developed.

Dual-energy computed tomography: pediatric considerations.

Multidetector computed tomography (CT) has revolutionized medicine and is now a fundamental aspect of modern radiology. Hardware and software advancements have significantly improved CT accessibility, image quality, and acquisition times. While considerable attention has been directed towards the potential risks of ionizing radiation from CT scans in children, recent concerns regarding the possible short- and long-term risks related to magnetic resonance imaging (MRI) conducted under general anesthesia have generated fresh interest in novel pediatric CT applications and techniques that allow imaging of awake patients at low radiation doses. Among these novel techniques, dual-energy CT (DECT) stands out for its ability to provide enhanced diagnostic information, reduce radiation doses further, and facilitate faster scans, making it a highly promising tool in pediatric radiology. This manuscript explores the current role of DECT in pediatric imaging, emphasizing its technical foundations, hardware configurations, and various reconstruction techniques. We discuss advanced post-processing techniques, such as material decomposition algorithms and virtual monoenergetic imaging, highlighting their clinical advantages in improving diagnostic accuracy and patient outcomes. Furthermore, the paper reviews the clinical applications of DECT in evaluating pulmonary perfusion, cardiovascular assessments, and oncologic imaging in pediatric patients.

Behavioral Experiments Online?

Online experiments offer several advantages over traditional laboratory experiments. However, for reaction time experiments, precise stimulus presentation and response detection is crucial. The precision of online experiments could be compromised due to increased variance arising from varying hardware configurations among participants, lack of control over experimental conditions, and the absence of an examiner. In this study, we conducted an online experiment using the avatar-Simon task to investigate whether small differences in reaction times could be examined using online experiments conducted with the experimental toolkit PsyToolkit. In the avatar-Simon task, participants respond to the color of vertically presented stimuli in front of avatars by pressing a left or right button. Reactions are faster when the position of the stimulus, defined from the avatar's point of view, matches the position of the response. Compared to the previous laboratory experiment, we observed lower effect sizes and more timeouts but were able to replicate the avatar-Simon effect overall. Based on further distributional and reliability analyses, PsyToolkit appears to be suitable tool to detect behavioral effects in the range of tens of milliseconds. We discuss differences and similarities with the original laboratory study and suggest how to address potential problems associated with online experiments.

A Robust Client Selection Mechanism for Federated Learning Environments

There is a exponential growth of data usage, specially due to the proliferation of connected applications with personalized models for different applications. In this context, Federated Learning (FL) emerges as a promising solution to enable collaborative model training while preserving the privacy and autonomy of participating clients. In a typical FL scenario, clients exhibit significant heterogeneity in terms of data distribution and hardware configurations. In this way, randomly sampling clients in each training round may not fully exploit the local updates from heterogeneous clients, resulting in lower model accuracy, slower convergence rate, degraded fairness, etc. In addition, malicious users could disseminate incorrect weights, which may decrease the accuracy of aggregated models and increase the time for convergence in FL. In this article, we introduce Resilience-aware Client Selection Mechanism for non-IID data and malicious clients in FL environment, called RICA. The proposed mechanism employs data size and entropy as criteria for client selection. In addition, RICA relies Centroid-Based Kernel Alignment (CKA) to identify and exclude potentially malicious clients. Our evaluation shows an improvement of 125% in Accuracy values in a scenario of malicious clients, which means the RICA+CKA demonstrates a more stable and resilient approach, reaching 90% accuracy in a few rounds compared to the default average approach, reached only around 30%. Therefore, results of the behavior of RICA+CKA in different datasets show the evaluation of different numbers of clients reaching around 90% while the other approach does not pass the 50% Accuracy.

Miyazaki Vermin Repulsion Robot and Its Adjustable Acousto-Optic Stimulus Generation Scheme

One of the most pressing issues in livestock farming is the protection of economically valuable livestock. The prevention and the treatment of infectious diseases are directly related to maintaining stable livestock output. Vermin is a primary source of livestock infection, resulting in the occurrence and expansion of epidemic diseases. To protect livestock against infections caused by epidemic diseases, this study proposes a vermin repulsion system called the Miyazaki Vermin Repulsion Robot (MiVeReR). Different from existing vermin repulsion systems, the development objective of MiVeReR is to repel vermin rather than kill them. In particular, MiVeReR generates changeable acousto-optic signals as repulsion signals for wild animals. Furthermore, MiVeReR employs image data to monitor the invasion of wild animals and their location data to track them, and accurately focuses the generated signals on them. These acousto-optic stimuli can be changed based on the reactions of the intruder through the feedback of the image data to ensure the effectiveness of the repulsion motions for vermin. Details on the hardware configuration of MiVeReR and its control scheme are explained. As a first step to develop MiVeReR, we attempted to repel vermin such as mice and wild cats from farm environments. Extensive experiments were conducted to verify the effectiveness of MiVeReR and the proposed control solution. Through experiments in wild environments, the feasibility of MiVeReR was inspected. The results of this study are concretely described.

Deploying large language models on diverse computing architectures: A performance evaluation framework

Deploying large language models (LLMs) across diverse computing architectures is a critical challenge in the field of artificial intelligence, particularly as these models become increasingly complex and resource-intensive. This review presents a performance evaluation framework designed to systematically assess the deployment of LLMs on various computing architectures, including CPUs, GPUs, TPUs, and specialized accelerators. The framework is structured around key performance metrics such as computational efficiency, latency, throughput, energy consumption, and scalability. It considers the trade-offs associated with different hardware configurations, optimizing the deployment to meet specific application requirements. The evaluation framework employs a multi-faceted approach, integrating both theoretical and empirical analyses to offer comprehensive insights into the performance dynamics of LLMs. This includes benchmarking LLMs under varying workloads, data batch sizes, and precision levels, enabling a nuanced understanding of how these factors influence model performance across different hardware environments. Additionally, the framework emphasizes the importance of model parallelism and distribution strategies, which are critical for efficiently scaling LLMs on high-performance computing clusters. A significant contribution of this framework is its ability to guide practitioners in selecting the optimal computing architecture for LLM deployment based on application-specific needs, such as low-latency inference for real-time applications or energy-efficient processing for large-scale deployments. The framework also provides insights into cost-performance trade-offs, offering guidance for balancing the financial implications of different deployment strategies with their performance benefits. Overall, this performance evaluation framework is a valuable tool for researchers and engineers, facilitating the efficient deployment of LLMs on diverse computing architectures. By offering a systematic approach to evaluating and optimizing LLM performance, the framework supports the ongoing development and application of these models across various domains. This paper will evaluate the deployment of large language models (LLMs) on diverse computing architectures, including x86, ARM, and RISC-V platforms. It will discuss strategies for optimizing LLM performance, such as dynamic frequency scaling, core scaling, and memory optimization. The research will contribute to understanding the best practices for deploying AI applications on different architectures, supporting technological innovation and competitiveness.

SWiLoc: Fusing Smartphone Sensors and WiFi CSI for Accurate Indoor Localization.

Dead reckoning is a promising yet often overlooked smartphone-based indoor localization technology that relies on phone-mounted sensors for counting steps and estimating walking directions, without the need for extensive sensor or landmark deployment. However, misalignment between the phone's direction and the user's actual movement direction can lead to unreliable direction estimates and inaccurate location tracking. To address this issue, this paper introduces SWiLoc (Smartphone and WiFi-based Localization), an enhanced direction correction system that integrates passive WiFi sensing with smartphone-based sensing to form Correction Zones. Our two-phase approach accurately measures the user's walking directions when passing through a Correction Zone and further refines successive direction estimates outside the zones, enabling continuous and reliable tracking. In addition to direction correction, SWiLoc extends its capabilities by incorporating a localization technique that leverages corrected directions to achieve precise user localization. This extension significantly enhances the system's applicability for high-accuracy localization tasks. Additionally, our innovative Fresnel zone-based approach, which utilizes unique hardware configurations and a fundamental geometric model, ensures accurate and robust direction estimation, even in scenarios with unreliable walking directions. We evaluate SWiLoc across two real-world environments, assessing its performance under varying conditions such as environmental changes, phone orientations, walking directions, and distances. Our comprehensive experiments demonstrate that SWiLoc achieves an average 75th percentile error of 8.89 degrees in walking direction estimation and an 80th percentile error of 1.12 m in location estimation. These figures represent reductions of 64% and 49%, respectively for direction and location estimation error, over existing state-of-the-art approaches.

GPAC-YOLOv8: lightweight target detection for fire scenarios

Abstract Due to the large number of parameters in the deep network model, it is difficult for existing fire detection methods to adapt to limited hardware configurations. In addition, detecting targets in the early stages of a fire is challenging owing to their small size. Therefore, this study presents a novel fire and smoke detection framework called GPAC-YOLOv8, which is based on the YOLOv8 architecture. Firstly, the integration of the ghost module and the Polarized Self-Attention attention mechanism into the backbone culminates in the CGP module, which is designed to improve computational efficiency while maintaining accuracy. Next, an innovative feature fusion module, AC-Neck, is developed through the application of the adaptive spatial feature fusion strategy and the lightweight content-aware reassembly of features upsampling mechanism, aiming to optimize feature map fusion and increase small target detection efficiency. Finally, a Focal-WIoU loss function, augmented with a dual weighting mechanism, is formulated to precisely delineate the aspect ratios of the predicted bounding boxes, thereby strengthening the generalization capacity of the model. Experimental results, derived from the application of the proposed GEAC-YOLOv8 method to a specially constructed dataset, show significant improvements in detection speed while maintaining detection accuracy compared to conventional methods. Thus, the GPAC-YOLOv8 framework demonstrably improves the effectiveness of object detection in fire and smoke scenarios.

Deep Learning Aided Intelligent Reflective Surfaces for 6G: A Survey

The envisioned sixth-generation (6G) networks anticipate robust support for diverse applications, including massive machine-type communications, ultra-reliable low-latency communications, and enhanced mobile broadband. Intelligent Reflecting Surfaces (IRS) have emerged as a key technology capable of intelligently reconfiguring wireless propagation environments, thereby enhancing overall network performance. Traditional optimization techniques face limitations in meeting the stringent performance requirements of 6G networks due to the intricate and dynamic nature of the wireless environment. Consequently, Deep Learning (DL) techniques are employed within the IRS framework to optimize wireless system performance. This paper provides a comprehensive survey of the latest research in DL-aided IRS models, covering optimal beamforming, resource allocation control, channel estimation and prediction, signal detection, and system deployment. The focus is on presenting promising solutions within the constraints of different hardware configurations. The survey explores challenges, opportunities, and open research issues in DL-aided IRS, considering emerging technologies such as digital twins (DTs), computer vision (CV), blockchain, network function virtualization (NFC), integrated sensing and communication (ISAC), software-defined networking (SDN), mobile edge computing (MEC), unmanned aerial vehicles (UAVs), and non-orthogonal multiple access (NOMA). Practical design issues associated with these enabling technologies are also discussed, providing valuable insights into the current state and future directions of this evolving field.

A national-wide survey on clinical implementation of PGx testing into precision therapeutics for Chinese children: a long way before standard clinical practice

BackgroundPharmacogenetics/pharmacogenomics (PGx) focuses on the genetic variation that causes the heterogeneity of pharmacokinetics and drug response among individuals and has the potential to predict individual efficacy and/or side effects. This study aims to investigate and understand the implementation of genetic testing for the personalized medication (GTPM) in children's hospitals in Mainland China.MethodsA survey was conducted on 50 children's hospitals from 31 provinces, municipalities, and autonomous regions across Mainland China, and statistical analysis and recommendations were made.ResultsQuestionnaire response was rate of 76.0% (38/50). Data from 15 hospitals conducting GTPM were included in this study, but only 6 hospitals had offered PGx tests for no less than five drug-related genes, and only 5 hospitals had covered more than ten drugs, which was a small scale overall. 20.0% of the laboratories did not conduct internal quality control, and 33.3% did not participate in inter-laboratory quality assessment. 46.7% of the practitioners did not receive external training. The primary goal for GTPM was to optimize drug dosage in the 15 hospitals, while the main challenge for GTPM was the implementation cost.ConclusionAlthough GTPM in pediatrics has made major progress in Mainland China in recent years, there were still various problems in terms of software, hardware configuration, personnel allocation, business scale, quality control, and result interpretation. This requires joint efforts of health administration, medical insurance departments, researchers, and hospitals to promote and improve GTPM.

A Custom-Developed Device for Testing Tensile Strength and Elasticity of Vascular and Intestinal Tissue Samples for Anastomosis Regeneration Research.

Optimizing the regeneration process of surgically created anastomoses (blood vessels, intestines, nerves) is an important topic in surgical research. One of the most interesting parameter groups is related to the biomechanical properties of the anastomoses. Depending on the regeneration process and its influencing factors, tensile strength and other biomechanical features may change during the healing process. Related to the optimal specimen size, the range and accuracy of measurements, and applicability, we have developed a custom-tailored microcontroller-based device. In this paper, we describe the hardware and software configuration of the latest version of the device, including experiences and comparative measurements of tensile strength and elasticity of artificial materials and biopreparate tissue samples. The machine we developed was made up of easily obtainable parts and can be easily reproduced on a low budget. The basic device can apply a force of up to 40 newtons, and can grasp a 0.05-1 cm wide, 0.05-1 cm thick tissue. The length of the test piece on the rail should be between 0.3 and 5 cm. Low production cost, ease of use, and detailed data recording make it a useful tool for experimental surgical research.

Comprehensive comparison between vision transformers and convolutional neural networks for face recognition tasks

This paper presents a comprehensive comparison between Vision Transformers and Convolutional Neural Networks for face recognition related tasks, including extensive experiments on the tasks of face identification and verification. Our study focuses on six state-of-the-art models: EfficientNet, Inception, MobileNet, ResNet, VGG, and Vision Transformers. Our evaluation of these models is based on five diverse datasets: Labeled Faces in the Wild, Real World Occluded Faces, Surveillance Cameras Face, UPM-GTI-Face, and VGG Face 2. These datasets present unique challenges regarding people diversity, distance from the camera, and face occlusions such as those produced by masks and glasses. Our contribution to the field includes a deep analysis of the experimental results, including a thorough examination of the training and evaluation process, as well as the software and hardware configurations used. Our results show that Vision Transformers outperform Convolutional Neural Networks in terms of accuracy and robustness against distance and occlusions for face recognition related tasks, while also presenting a smaller memory footprint and an impressive inference speed, rivaling even the fastest Convolutional Neural Networks. In conclusion, our study provides valuable insights into the performance of Vision Transformers for face recognition related tasks and highlights the potential of these models as a more efficient solution than Convolutional Neural Networks.

A Global Feature Reused Network for Defect Detection in Steel Images

Abstract Accurate detection of surface defects for steel is essential to improve surface quality and service life. Deep learning (DL) used in steel surface defect detection can solve the problems of low efficiency and poor accuracy of traditional manual detection. The classic YOLOv5 as a DL method is used to accomplish defect detection tasks without attention mechanisms, resulting in a loss of global information. Besides, it is difficult to complete complex network detection tasks with low-configuration hardware, especially for surface defects with complex defect types and variable defect sizes. To solve these issues, this paper introduces an improved global feature reuse and hardware-aware YOLOv5 by using BoTNet, RepGhost, and EfficientRep model (BGE-YOLOv5). The multi-head self-attention layer is used to obtain global information and only part of the convolutional layers is replaced to avoid the excessive computational cost. The RepGhost model is introduced to extract the remaining feature information for feature reuse. EfficientRep is used to replace the original structure to achieve hardware-aware and to balance the detection veracity and efficiency. The distance IOU is replaced by SCYLLA-IOU to accelerate the iteration and improve stability. The results of the framework on the surface defect database (NEU-DET) show that BGE-YOLOv5 achieves a mean average precision of 79.5%, which is 10.3% greater than the baseline. The proposed BGE-YOLOv5 has a better performance in steel surface defect detection.

Using in situ research-based design to explore learning module effectiveness and usability in a virtual reality system for workforce training

This study examines the implementation and effectiveness of Virtual Reality (VR) into employee communication training for Texas Department of Transportation employees. The study also explores the impact of iterative, in situ research-based design on VR system usability, and potential relationships between the usability and ergonomics of VR devices, and user receptiveness to VR content. For this work we adapt the definition of in situ to mean a learning activity that takes place in the location or environment where the participants will actually be trained. Following the pandemic, many office-based workers adopted hybrid-remote work formats (Yang, Kim, & Hong, 2023), necessitating updated approaches to employee training and instruction which VR may provide. Further, VR may offer opportunities for private role-play for employee communication practice, as well as an overall more high-fidelity learning opportunity. Findings indicate that VR is an effective and engaging solution for workforce and management communication training, and that hardware configurations influence overall employee enthusiasm for VR-based training.

Real-Time Surveillance with AI: A Comprehensive Approach to Security and Monitoring

The last ten years have been quite phenomenal for artificial intelligence and computer vision. For the past few years, it was unstoppable, with outstanding breakthroughs that moved towards advanced real-time surveillance systems-that wholesome approach to real-time surveillance through the power of AI in technological advancement for accuracy, scalability, and responsiveness. The system integrates facial recognition, object detection, and anomaly detection algorithms to give a very stable, adaptive response to the various facets of surveillance needs, such as public safety, institutional monitoring, and private security. Through deep learning models extracted from comprehensive datasets, the system is able to identify known individuals in real time and trace the movements and unusual behaviors of these people. A discussion of challenges in data privacy, computational efficiency, and false positives goes hand in hand with an in-depth analysis of system architecture, which includes thorough hardware and software configuration descriptions. The experimental results show the potential of the system in real-world environments as an alternate scalable solution for future security infrastructures.

FPGA-Based Sparse Matrix Multiplication Accelerators: From State-of-the-art to Future Opportunities

Sparse matrix multiplication (SpMM) plays a critical role in high-performance computing applications, such as deep learning, image processing, and physical simulation. Field-Programmable Gate Arrays (FPGAs), with their configurable hardware resources, can be tailored to accelerate SpMMs. There has been considerable research on deploying sparse matrix multipliers across various FPGA platforms. However, the FPGA-based design of sparse matrix multipliers still presents numerous challenges. Therefore, it is necessary to summarize and organize the current work to provide a reference for further research. This paper first introduces the computational method of SpMM, and categorizes the different challenges of FPGA deployment. Following this, we introduce and analyze a variety of state-of-the-art FPGA-based accelerators tailored for SpMMs. In addition, a comparative analysis of these accelerators is performed, examining metrics including compression rate, throughput, and resource utilization. Finally, we propose potential research directions and challenges for further study of FPGA-based SpMM acclerators.