Proposed Architecture Research Articles

Semi-Supervised Medical Image Segmentation Using Cross-Style Consistency With Shape-Aware and Local Context Constraints.

Despite the remarkable progress in semi-supervised medical image segmentation methods based on deep learning, their application to real-life clinical scenarios still faces considerable challenges. For example, insufficient labeled data often makes it difficult for networks to capture the complexity and variability of the anatomical regions to be segmented. To address these problems, we design a new semi-supervised segmentation framework that aspires to produce anatomically plausible predictions. Our framework comprises two parallel networks: shape-agnostic and shape-aware networks. These networks learn from each other, enabling effective utilization of unlabeled data. Our shape-aware network implicitly introduces shape guidance to capture shape fine-grained information. Meanwhile, shape-agnostic networks employ uncertainty estimation to further obtain reliable pseudo-labels for the counterpart. We also employ a cross-style consistency strategy to enhance the network's utilization of unlabeled data. It enriches the dataset to prevent overfitting and further eases the coupling of the two networks that learn from each other. Our proposed architecture also incorporates a novel loss term that facilitates the learning of the local context of segmentation by the network, thereby enhancing the overall accuracy of prediction. Experiments on three different datasets of medical images show that our method outperforms many excellent semi-supervised segmentation methods and outperforms them in perceiving shape. The code can be seen at https://github.com/igip-liu/SLC-Net.

Federated Learning for Digital Twin-Based Vehicular Networks: Architecture and Challenges

A digital twin uses a virtual model of the physical system to fulfill the diverse requirements (e.g., latency, reliability, quality of physical experience) for emerging vehicular network applications. Although a twin-based implementation of vehicular networks can offer performance optimization, modeling a digital twin is a significantly challenging task. Federated learning (FL) is a better privacy-preserving, distributed learning scheme that can be used to model twin models. Although FL can offer performance enhancement, it requires careful design. Therefore, in this article, we present an overview of FL for a twinbased vehicular network. A general architecture for FL-enabled digital twins for a vehicular network is presented. Our proposed architecture consists of two spaces, such as twin space and the physical space. A physical space consists of all the physical entities (e.g., cars and edge servers) required for vehicular networks, whereas, the twin space refers to the logical space that is used for the deployment of twins. A twin space can be implemented either using edge servers and cloud servers. We also outline a few use cases of FL for the twin-based vehicular network. Finally, the paper is concluded and an outlook on open challenges is presented.

DAST-Net: Dense visual attention augmented spatio-temporal network for unsupervised video anomaly detection

This paper introduces an innovative end-to-end trainable framework named Dense Attention-aware Spatio-Temporal Network (DAST-Net) for video anomaly detection. The framework adeptly leverages both spatial and temporal data in an unsupervised manner, eliminating the need for manually crafted features. To enhance spatial feature representation, DAST-Net incorporates visual attention-aware residual connections within the denser residual network (DenserResNet), deviating from traditional identity skip connections. The rationale behind this connection choice is to augment the contextual understanding of features across various scales. For capturing temporal patterns, the framework employs a Convolutional LSTM Autoencoder (ConvLSTM-AE) module, enabling effective learning and representation of temporal dependencies in video data. Consequently, discriminating features from attention modules are combined with the features extracted by the ConvLSTM-AE module, enhancing visual recognition capabilities for both spatial and temporal aspects. Our proposed architecture outperforms state-of-the-art methods on four benchmark datasets, showcasing AUC scores of 85.4% on Ped1, 97.9% on Ped2, 89.8% on Avenue, and 73.7% on the ShanghaiTech dataset. The results demonstrate the performance of our method in identifying unusual events in video data.

An edge architecture for enabling autonomous aerial navigation with embedded collision avoidance through remote nonlinear model predictive control

In this article, we present an edge-based architecture for enhancing the autonomous capabilities of resource-constrained aerial robots by enabling a remote nonlinear model predictive control scheme, which can be computationally heavy to run on the aerial robots' onboard processors. The nonlinear model predictive control is used to control the trajectory of an unmanned aerial vehicle while detecting, and preventing potential collisions. The proposed edge architecture enables trajectory recalculation for resource-constrained unmanned aerial vehicles in relatively real-time, which will allow them to have fully autonomous behaviors. The architecture is implemented with a remote Kubernetes cluster on the edge side, and it is evaluated on an unmanned aerial vehicle as our controllable robot, while the robotic operating system is used for managing the source codes, and overall communication. With the utilization of edge computing and the architecture presented in this work, we can overcome computational limitations, that resource-constrained robots have, and provide or improve features that are essential for autonomous missions. At the same time, we can minimize the relative travel time delays for time-critical missions over the edge, in comparison to the cloud. We investigate the validity of this hypothesis by evaluating the system's behavior through a series of experiments by utilizing either the unmanned aerial vehicle or the edge resources for the collision avoidance mission.

Adaptively Placed Multi-Grid Scene Representation Networks for Large-Scale Data Visualization.

Scene representation networks (SRNs) have been recently proposed for compression and visualization of scientific data. However, state-of-the-art SRNs do not adapt the allocation of available network parameters to the complex features found in scientific data, leading to a loss in reconstruction quality. We address this shortcoming with an adaptively placed multi-grid SRN (APMGSRN) and propose a domain decomposition training and inference technique for accelerated parallel training on multi-GPU systems. We also release an open-source neural volume rendering application that allows plug-and-play rendering with any PyTorch-based SRN. Our proposed APMGSRN architecture uses multiple spatially adaptive feature grids that learn where to be placed within the domain to dynamically allocate more neural network resources where error is high in the volume, improving state-of-the-art reconstruction accuracy of SRNs for scientific data without requiring expensive octree refining, pruning, and traversal like previous adaptive models. In our domain decomposition approach for representing large-scale data, we train an set of APMGSRNs in parallel on separate bricks of the volume to reduce training time while avoiding overhead necessary for an out-of-core solution for volumes too large to fit in GPU memory. After training, the lightweight SRNs are used for realtime neural volume rendering in our open-source renderer, where arbitrary view angles and transfer functions can be explored. A copy of this paper, all code, all models used in our experiments, and all supplemental materials and videos are available at https://github.com/skywolf829/APMGSRN.

Continuous Sign Language Recognition Using Combination of Two Stream 3DCNN and SubUNet

Research on sign language recognition using deep learning has been carried out by many researchers in the field of computer science but there are still obstacles in achieving the expected level of accuracy. Not a few researchers who want to do research for Continuous Sign Language Recognition but are trapped into research for Isolated Sign Language Recognition. The purpose of this study was to find the best method for performing Continuous Sign Language Recognition using Deep Learning. The 2014 RWTH-PHOENIX-Weather dataset was used in this study. The dataset was obtained from a literature study conducted to find datasets that are commonly used in Continuous Sign Language Recognition research. The dataset is used to develop the proposed method. The combination of 3DCNN, LSTM and CTC models is used to form part of the proposed method architecture. The collected dataset is also converted into an Optical Flow frame sequence to be used as Two Stream input along with the original RGB frame sequence. Word Error Rate on the prediction results is used to review the performance of the developed method. Through this research, the best achieved Word Error Rate is 94.1% using the C3D BLSTM CTC model with spatio stream input.

Architecture design of a vehicle–road-cloud collaborative automated driving system

Transportation systems serve as a crucial foundation for maintaining the normal operation of cities and satisfying the requirements of public life. With the development of next-generation information technologies, automated driving technologies have brought new opportunities to improve the performance of traffic systems and the intelligence level of cities. Currently, significant research efforts have been conducted to develop automated driving systems in three major industries, i.e., automobile, roadway, and telecommunication. However, the collaboration and integration of automated driving systems among automobile, roadway, and telecommunications are still lacking, especially for collaborative development of system architecture and objectives. To address the need, this study first proposes a system architecture of vehicle–road-cloud collaborative automated driving system (VRC-CADS). Three levels of collaborative development, i.e., collaborative sensing, collaborative decision-making, and collaborative control, are designed for the VRC-CADS. Based on that, the typical scenarios of automated driving for each level of the system are further defined and interpreted. Moreover, feasible and systematic suggestions for the collaborative development of the VRC-CADS are provided, considering the cross-cutting collaboration among government agencies, academia, and industry. The proposed system architecture of the VRC-CADS will facilitate the optimization of urban lifelines and the evolution of intelligent cities.

Automated Detection of Persuasive Content in Electronic News

Persuasive content in online news contains elements that aim to persuade its readers and may not necessarily include factual information. Since a news article only has some sentences that indicate persuasiveness, it would be quite challenging to differentiate news with or without the persuasive content. Recognizing persuasive sentences with a text summarization and classification approach is important to understand persuasive messages effectively. Text summarization identifies arguments and key points, while classification separates persuasive sentences based on the linguistic and semantic features used. Our proposed architecture includes text summarization approaches to shorten sentences without persuasive content and then using classifiers model to detect those with persuasive indication. In this paper, we compare the performance of latent semantic analysis (LSA) and TextRank in text summarization methods, the latter of which has outperformed in all trials, and also two classifiers of convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM). We have prepared a dataset (±1700 data and manually persuasiveness-labeled) consisting of news articles written in the Indonesian language collected from a nationwide electronic news portal. Comparative studies in our experimental results show that the TextRank–BERT–BiLSTM model achieved the highest accuracy of 95% in detecting persuasive news. The text summarization methods were able to generate detailed and precise summaries of the news articles and the deep learning models were able to effectively differentiate between persuasive news and real news.

DTAN: Diffusion-based Text Attention Network for medical image segmentation

In the current era, diffusion models have emerged as a groundbreaking force in the realm of medical image segmentation. Against this backdrop, we introduce the Diffusion Text-Attention Network (DTAN), a pioneering segmentation framework that amalgamates the principles of text attention with diffusion models to enhance the precision and integrity of medical image segmentation. Our proposed DTAN architecture is designed to steer the segmentation process towards areas of interest by leveraging a text attention mechanism. This mechanism is adept at identifying and zeroing in on the regions of significance, thus improving the accuracy and robustness of the segmentation. In parallel, the integration of a diffusion model serves to diminish the influence of noise and irrelevant background data in medical images, thereby improving the quality of the segmentation results. The diffusion model is instrumental in filtering out extraneous factors, allowing the network to more effectively capture the nuances and characteristics of the target regions, which in turn enhances segmentation precision. We have subjected DTAN to rigorous evaluation across three datasets: Kvasir-Sessile, Kvasir-SEG, and GlaS. Our focus was particularly drawn to the Kvasir-Sessile dataset due to its relevance to clinical applications. When benchmarked against other state-of-the-art methods, our approach demonstrated significant improvements on the Kvasir-Sessile dataset, with a 2.77% increase in mean Intersection over Union (mIoU) and a 3.06% increase in mean Dice Similarity Coefficient (mDSC). These results provide strong evidence of the DTAN’s generalizability and robustness, and its distinct advantages in the task of medical image segmentation.

End-to-end system for monitoring the state of rivers using a drone

Automatic data analysis systems in the Internet of Things are a key element. One such case is the use of drones to monitor rivers, which are quite often located around built-up areas. This is an important element for the analysis of urban areas that are exposed to various environmental challenges such as pollution and animal habitats. Data analysis allows the detection of anomalies in the state of rivers, reducing the risk of ecological disasters or even floods. Additionally, constant control of areas enables analysis of the impact of urbanization on a given area as well as environmental protection. In this paper, we propose an end-to-end system, where the user performs measurements with a drone and the result is a segmentation mask from the U-Net network, but improved by image processing algorithms. The system is based on performing segmentation with a neural network, imposing the obtained mask on the image that was previously subjected to edge detection. All pixels under the mask are analyzed by the clustering method in terms of belonging to a river or bank. In addition, when there are other measurements from the same area, they are used to compare and analyze changes. The proposed system architecture is based on the automation of activities due to the combination of various graphics processing methods. Moreover, the method allows for obtaining more accurate segmentation results than classic methods. The proposition was tested on data gathered near river areas in southern Poland to show the possibilities and effectiveness of the system. Proposed methodology reached 0.8524 of Dice coefficient using VGG16 as encoder.

COMFormer: Classification of Maternal-Fetal and Brain Anatomy using a Residual Cross-Covariance Attention Guided Transformer in Ultrasound.

Monitoring the healthy development of a fetus requires accurate and timely identification of different maternal-fetal structures as they grow. To facilitate this objective in an automated fashion, we propose a deep-learning-based image classification architecture called the COMFormer to classify maternal-fetal and brain anatomical structures present in two-dimensional fetal ultrasound images. The proposed architecture classifies the two subcategories separately: maternal-fetal (abdomen, brain, femur, thorax, mother's cervix, and others) and brain anatomical structures (trans-thalamic, trans-cerebellum, trans-ventricular, and non-brain). Our proposed architecture relies on a transformer-based approach that leverages spatial and global features by using a newly designed residual cross-variance attention (R-XCA) block. This block introduces an advanced cross-covariance attention mechanism to capture a long-range representation from the input using spatial (e.g., shape, texture, intensity) and global features. To build COMFormer, we used a large publicly available dataset (BCNatal) consisting of 12, 400 images from 1,792 subjects. Experimental results prove that COMFormer outperforms the recent CNN and transformer-based models by achieving 95.64% and 96.33% classification accuracy on maternal-fetal and brain anatomy, respectively.

Efficient Hardware Implementation of Elliptic-Curve Diffie–Hellman Ephemeral on Curve25519

Hardware architecture optimized for implementing the elliptic-curve Diffie–Hellman ephemeral (ECDHE) on 256-bit Montgomery elliptic curves presents unique challenges, particularly for resource-constrained IoT and mobile devices. This work aims to provide an efficient hardware implementation of ECDHE on Curve25519, including a dedicated finite state machine (FSM) designed to handle point multiplication and ECDHE operations, utilizing constant-time algorithms and a unified memory block for resource management. Additionally, we introduce an optimized modular computation unit that covers modular addition, subtraction, multiplication, and inversion. Our proposed hardware architecture enhances the efficiency of ECDHE operations while maintaining low resource utilization, considerably reduced latency, and low power consumption. Synthesized on the Xilinx Artix-7 platform, our design boasts 64,000 Slices and a clock speed of 102 MHz, and it computes an ECDHE scalar multiplication operation in 1.1 ms, consuming 117 mW. The proposed hardware design can be applied to various platforms, including mobile devices and IoT systems.

Event-Triggered Control with Intermittent Communications over Erasure Channels for Leader–Follower Problems with the Combined-Slip Effect

In this article, we investigate the vehicle path-following problem for a vehicle-to-vehicle (V2V)–enabled leader–follower scenario and propose an integrated control policy for the following vehicle to accurately follow the leader’s path. We propose a control strategy for the follower vehicle to maintain a velocity-dependent distance relative to the leader vehicle while stabilizing its longitudinal and lateral dynamics considering the combined-slip effect and tire force saturation. In light of reducing wireless communication errors and efficient usage of battery power and resources, we propose an intermittent V2V communication in which transmissions are scheduled based on an event-triggered law. An event is triggered and a transmission is scheduled in subsequent sample time if some of the well-defined path-following error functions (relative distance error and lateral error) exceed given tolerance bounds. Considering that the V2V communication channel might be erroneous or a transmission fails due to, e.g., vehicles’ distance or low battery power, we consider data loss in the V2V channel. Our proposed control law consists of two components: a receding horizon feedback controller with state constraints based on a safe operation envelop and a feedforward controller that generates complementary control inputs when the leader’s states are successfully communicated to the follower. To mitigate the effects of data loss on the follower’s path-following performance, we design a remote estimator for the follower to predict the leader’s state using its on-board sensor equipment when an event is triggered but the corresponding state information is not received by the follower due to a packet loss. Incorporating this estimator allows the follower to apply cautionary control inputs knowing that the path-following error had exceeded a tolerance bound. We show that while the feedback controller stabilizes the follower’s dynamics, the feedforward component improves the safety margins and reduces the path-following errors even in the presence of data loss. High-fidelity simulations are performed using CarSim to validate the effectiveness of our proposed control architecture specifically in harsh maneuvers and high-slip scenarios on various road surface conditions.

Enhancing Cable Gland Design Efficiency through TRIZ-Driven Innovation

This study introduces a novel approach to cable gland design through the utilization of TRIZ tools. A cable gland serves two primary functions: retaining and sealing cables within an enclosure, while also offering secondary features such as earthing and cable protection. The number of components in a gland type is determined by functional requirements. Given its classification as a low-cost, high-volume product, the competitive edge lies in achieving reduced selling costs and simplified installation. Addressing key challenges, which include a high component count (currently nine) and a cumbersome installation process, forms the core focus of this work. The research outlines the TRIZ-guided process and methodology employed to devise inventive solutions to these issues, resulting in a component reduction from nine to four. Moreover, the study encompasses the incorporation of various constraints in the development of efficient cable gland designs, yielding several innovative concepts. The paper also introduces proposed system designs and architectures, established based on TRIZ-derived inventive principles, paving the way for subsequent system prototype development. The work underscores the accelerated problem identification and solution generation facilitated by TRIZ tools. The evolving concept is presently situated within the feasibility and design phase, effectively addressing both primary and secondary challenges through TRIZ methodologies. Keywords: Inventive Problem solving, TRIZ, Engineering Problem, Cabled land, Trimming

Research on carbon emission prediction and economic policy based on TCN-LSTM combined with attention mechanism

IntroductionIn the face of increasingly severe global climate change and environmental challenges, reducing carbon emissions has become a key global priority. Deep learning, as a powerful artificial intelligence technology, has demonstrated significant capabilities in time series analysis and pattern recognition, opening up new avenues for carbon emission prediction and policy development.MethodsIn this study, we carefully collected and pre-processed four datasets to ensure the reliability and consistency of the data. Our proposed TCN-LSTM combination architecture effectively leverages the parallel computing capabilities of TCN and the memory capacity of LSTM, more efficiently capturing long-term dependencies in time series data. Furthermore, the introduction of an attention mechanism allows us to weigh important factors in historical data, thereby improving the accuracy and robustness of predictions. ResultsOur research findings provide novel insights and methods for advancing carbon emission prediction. Additionally, our discoveries offer valuable references for decision-makers and government agencies in formulating scientifically effective carbon reduction policies. As the urgency of addressing climate change continues to grow, the progress made in this paper can contribute to a more sustainable and environmentally conscious future. DiscussionIn this paper, we emphasize the potential of deep learning techniques in carbon emission prediction and demonstrate the effectiveness of the TCN-LSTM combination architecture. The significant contribution of this research lies in providing a new approach to address the carbon emission prediction problem in time series data. Moreover, our study underscores the importance of data reliability and consistency for the successful application of models. We encourage further research and application of this method to facilitate the achievement of global carbon reduction goals.

Numerical Simulation and Design of Ensemble Learning Based Improved Software Development Effort Estimation System

This research paper proposes a novel approach to improving software development effort estimation by integrating ensemble learning algorithms with numerical simulation techniques. The objective of this study is to design an ensemble learning-based software development effort estimation system that leverages the strengths of multiple algorithms to enhance accuracy and reliability. The proposed system combines the power of ensemble learning, which involves aggregating predictions from multiple models, with numerical simulation techniques that enable the modelling and analysis of complex software development processes. A diverse set of software development projects is collected, encompassing various domains, sizes, and complexities. Ensemble learning algorithms such as Random Forest, Gradient Boosting, Bagging, and AdaBoost are selected for their ability to capture different aspects of the data and produce robust predictions. The proposed system architecture is presented, illustrating the flow of data and components. A model training and evaluation pipeline is developed, enabling the integration of ensemble learning and numerical simulation modules. The system combines the predictions generated by the ensemble models with the simulation results to produce more accurate and reliable effort estimates. The experimental setup involves a comprehensive evaluation of the proposed system. A real-world dataset comprising historical project data is utilized, and various performance metrics, including Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), are employed to assess the effectiveness of the system. The results and analysis demonstrate that the ensemble learning-based effort estimation system outperforms traditional techniques, showcasing its potential to enhance project planning and resource allocation.

A model of service-oriented architecture of e-certification system to support boat registration and site visit inspection to support maritime safety and crew health inspection

AbstractThis article describes the architecture of software services that provide registration of small boat data to build maritime safety. The proposed system architecture which so-called SIMKAPEL aims to support small boat registration for maritime safety. It is developed from a conventional service-oriented architecture (SOA) with an open standard for data interoperability between elements. The goal of this research is to implement and evaluate the proposed architecture to support small boat registration and e-certification. Several parts of the architecture have been implemented as a platform that provides core services and data sources to be used by Ministry of Transportation, Directorate General of Sea Transportation, and other government agencies. Finally, the proposed SIMKAPEL platform with the SOA architecture is evaluated for its benefits to users through the expert opinion survey to measure its efficiency. The analysis results showed that SOA-based SIMKAPEL is feasible to use as a boat e-certification system. Furthermore, experiment results provide three parameters (e.g., maritime safety, namely, cargo/passengers, boat safety facilities, and maritime safety systems) with efficiency values of 42%, 46%, and 42%, respectively. The parameter with the highest efficiency shows a big effect on the level of safety of the boat and passengers. Thus, the parameters that need to be prioritized and need boat owners’ attention are cargo/passenger, boat safety equipment, and maritime safety system.

Harnessing optical flow in deep learning framework for cardiopulmonary resuscitation training

With the incidence of out-of-hospital cardiac arrest increasing on a global scale, cardiopulmonary resuscitation (CPR) has become crucial in maximizing the survival rate of cardiac arrest patients. However, the feasibility of widespread CPR training remains limited at present, in part owing to the high costs required to equip manikins with feedback functions. We therefore propose a deep learning technology designed to provide a low-cost CPR education. Our model converts chest compression (CC) videos captured via smartphone into appropriate images, and provides feedback on four core components of high-quality CPR: CC count, maximum compression depth of CCs, complete chest release following compression, and incorrect hand positions of CCs. To reduce the complexity of video processing, we conducted a composite-image-based evaluation for time-independent tasks. Our proposed architecture achieved satisfactory performance on all tasks, demonstrating high cost-effectiveness. Moreover, by focusing on educational context, our approach helps expand educational opportunities and enhance CPR training. By allowing many individuals to simultaneously receive CPR training in an optimized educational environment, we can ultimately improve the survival rate of cardiac arrest patients.

A deep learning feed-forward neural network framework for the solutions to singularly perturbed delay differential equations

In this paper, we explore a deep learning feedforward artificial neural network (ANN) framework as a numerical tool for approximating the solutions to singularly perturbed delay differential equations (SPDDE). Our approach trains the network with fewer uniform data points along with linear interpolation as a dependent variable. More importantly, the exact solution is not used for training the network. A mean square or Euclidean norm type total loss function is utilized to assess the deep learning network’s performance. In the training and testing phase, our network adjusts itself to fit the tangent and the curvature by minimizing the total loss function and fine-tuning the network’s hyperparameters during the backpropagation stage. Our focus in this contribution is to investigate an answer to the question- “whether neural networks can learn to solve the differential equations with both delay and boundary layer behavior”? Traditional numerical methods are known to perform poorly in approximating the solutions to SPDDE due to boundary layer behavior, which is the sharp change in the gradient of the solution inside a region of very small width. Our proposed network architecture is used to investigate the above question for three varieties of SPDDE. The numerical results demonstrate that the fine-tuned adaptive deep learning architecture can effectively approximate the solution to SPDDE for varieties of delay and perturbation parameters. It is expected that the current method can be extendable to approximate 2D convection-diffusion partial differential equations with a boundary layer.

Design and implementation of IMU-based locomotion mode recognition system on Zynq SoC

Active prostheses are becoming an increasingly viable option for lower limb amputees, as they can significantly improve their quality of life and mobility. However, to ensure a robust and effective control of these prostheses, the terrain environment must be considered. In this study, we have proposed a hardware implementation of a real-time low-latency human Locomotion Mode Recognition (LMR) system on a system on chip (SoC) platform from Xilinx, which can be customized and optimized based on the user's needs without requiring external computing resources or human intervention. The system uses a single IMU sensor placed on the shank, and includes all the stages in a LMR process, i.e. signal pre-processing, feature extraction, and terrains estimation using a MLP neural network classifier to classify five terrains (level walking, stair ascent/descent, ramp ascent/descent). To achieve a flexible and efficient hardware design, the proposed system architecture was optimized using parallelism and quantization optimization techniques. This approach has resulted in a significant performance improvement, with a processing speed 15 times faster than the PL non-optimized approach, and 4.3 times faster than the PS(-O3) optimized implementation on the same Zynq FPGA device. The proposed architecture was also validated in real time via the analog discovery device.