Search Results (6,275)

In this research, a multi-channel target speech enhancement scheme is proposed that is based on deep learning (DL) architecture and assisted by multi-source tracking using a labeled random finite set (RFS) framework. A neural network based on minimum variance distortionless response (MVDR) beamformer is considered as the beamformer of choice, where a residual dense convolutional graph-U-Net is applied in a generative adversarial network (GAN) setting to model the beamformer for target speech enhancement under reverberant conditions involving multiple moving speech sources. The input dataset for this neural architecture is constructed by applying multi-source tracking using multi-sensor generalized labeled multi-Bernoulli (MS-GLMB) filtering, which belongs to the labeled RFS framework, to obtain estimations of the sources’ positions and the associated labels (corresponding to each source) at each time frame with high accuracy under the effect of undesirable factors like reverberation and background noise. The tracked sources’ positions and associated labels help to correctly discriminate the target source from the interferers across all time frames and generate time–frequency (T-F) masks corresponding to the target source from the output of a time-varying, minimum variance distortionless response (MVDR) beamformer. These T-F masks constitute the target label set used to train the proposed deep neural architecture to perform target speech enhancement. The exploitation of MS-GLMB filtering and a time-varying MVDR beamformer help in providing the spatial information of the sources, in addition to the spectral information, within the neural speech enhancement framework during the training phase. Moreover, the application of the GAN framework takes advantage of adversarial optimization as an alternative to maximum likelihood (ML)-based frameworks, which further boosts the performance of target speech enhancement under reverberant conditions. The computer simulations demonstrate that the proposed approach leads to better target speech enhancement performance compared with existing state-of-the-art DL-based methodologies which do not incorporate the labeled RFS-based approach, something which is evident from the 75% ESTOI and PESQ of 2.70 achieved by the proposed approach as compared with the 46.74% ESTOI and PESQ of 1.84 achieved by Mask-MVDR with self-attention mechanism at a reverberation time (RT60) of 550 ms. Full article

The Euler-Type Universal Numerical Integrator (E-TUNI) is a discrete numerical structure that couples a first-order Euler-type numerical integrator with some feed-forward neural network architecture. Thus, E-TUNI can be used to model non-linear dynamic systems when the real-world plant’s analytical model is unknown. From the discrete solution provided by E-TUNI, the integration process can be either forward or backward. Thus, in this article, we intend to use E-TUNI in a backward integration framework to model autonomous non-linear dynamic systems. Three case studies, including the dynamics of the non-linear inverted pendulum, were developed to verify the computational and numerical validation of the proposed model. Full article

The deep learning architecture YOLO (You Only Look Once) has demonstrated its superior visual detection performance in various computer vision tasks and has been widely applied in the field of automatic surface defect detection. In this paper, we propose a lightweight YOLOv8-based method for the quality inspection of car body welding spots. We developed a TA-YOLOv8 network structure which has an improved Task-Aligned (TA) head detection, designed to handle a small sample size, imbalanced positive and negative samples, and high-noise characteristics of Body-in-White welding spot data. By learning with fewer parameters, the model achieves more efficient and accurate classification. Additionally, our algorithm framework can perform anomaly segmentation and classification on our open-world raw datasets obtained from actual production environments. The experimental results show that the lightweight module improves the processing speed by an average of 2.8%, with increases in detection the mAP@50-95 and recall rate of 1.35% and 0.1226, respectively. Full article

Pedestrians, as the most vulnerable road users in traffic crashes, prompt transportation researchers and urban planners to prioritize pedestrian safety due to the elevated risk and growing incidence of injuries and fatalities. Thorough pedestrian crash data are indispensable for safety research, as the most detailed descriptions of crash scenes and pedestrian actions are typically found in crash narratives and diagrams. However, extracting and analyzing this information from police crash reports poses significant challenges. This study tackles these issues by introducing innovative image-processing techniques to analyze crash diagrams. By employing cutting-edge technological methods, the research aims to uncover and extract hidden features from pedestrian crash data in Michigan, thereby enhancing the understanding and prevention of such incidents. This study evaluates the effectiveness of three Convolutional Neural Network (CNN) architectures—VGG-19, AlexNet, and ResNet-50—in classifying multiple hidden features in pedestrian crash diagrams. These features include intersection type (three-leg or four-leg), road type (divided or undivided), the presence of marked crosswalk (yes or no), intersection angle (skewed or unskewed), the presence of Michigan left turn (yes or no), and the presence of nearby residentials (yes or no). The research utilizes the 2020–2023 Michigan UD-10 pedestrian crash reports, comprising 5437 pedestrian crash diagrams for large urbanized areas and 609 for rural areas. The CNNs underwent comprehensive evaluation using various metrics, including accuracy and F1-score, to assess their capacity for reliably classifying multiple pedestrian crash features. The results reveal that AlexNet consistently surpasses other models, attaining the highest accuracy and F1-score. This highlights the critical importance of choosing the appropriate architecture for crash diagram analysis, particularly in the context of pedestrian safety. These outcomes are critical for minimizing errors in image classification, especially in transportation safety studies. In addition to evaluating model performance, computational efficiency was also considered. In this regard, AlexNet emerged as the most efficient model. This understanding is precious in situations where there are limitations on computing resources. This study contributes novel insights to pedestrian safety research by leveraging image processing technology, and highlights CNNs’ potential use in detecting concealed pedestrian crash patterns. The results lay the groundwork for future research, and offer promise in supporting safety initiatives and facilitating countermeasures’ development for researchers, planners, engineers, and agencies. Full article

The incorporation of artificial intelligence (AI) into sixth-generation (6G) mobile networks is expected to revolutionize communication systems, transforming them into intelligent platforms that provide seamless connectivity and intelligent services. This paper explores the evolution of 6G architectures, as well as the enabling technologies required to integrate AI across the cloud, core network (CN), radio access network (RAN), and terminals. It begins by examining the necessity of embedding AI into 6G networks, making it a native capability. The analysis then outlines potential evolutionary paths for the RAN architecture and proposes an end-to-end AI-driven framework. Additionally, key technologies such as cross-domain AI collaboration, native computing, and native security mechanisms are discussed. The study identifies potential use cases, including embodied intelligence, wearable devices, and generative AI, which offer valuable insights into fostering collaboration within the AI-driven ecosystem and highlight new revenue model opportunities and challenges. The paper concludes with a forward-looking perspective on the convergence of AI and 6G technology. Full article

This paper explores the application of the ESP32 microcontroller in edge computing, focusing on the design and implementation of an edge server system to evaluate performance improvements achieved by integrating edge and cloud computing. Responding to the growing need to reduce cloud burdens and latency, this research develops an edge server, detailing the ESP32 hardware architecture, software environment, communication protocols, and server framework. A complementary cloud server software framework is also designed to support edge processing. A deep learning model for object recognition is selected, trained, and deployed on the edge server. Performance evaluation metrics, classification time, MQTT (Message Queuing Telemetry Transport) transmission time, and data from various MQTT brokers are used to assess system performance, with particular attention to the impact of image size adjustments. Experimental results demonstrate that the edge server significantly reduces bandwidth usage and latency, effectively alleviating the load on the cloud server. This study discusses the system’s strengths and limitations, interprets experimental findings, and suggests potential improvements and future applications. By integrating AI and IoT, the edge server design and object recognition system demonstrates the benefits of localized edge processing in enhancing efficiency and reducing cloud dependency. Full article

Pattern recognition applications involve extensive arithmetic operations, including additions, multiplications, and divisions. When implemented on resource-constrained edge devices, these operations demand dedicated hardware, with division being the most complex. Conventional hardware dividers, however, incur substantial overhead in terms of resource consumption and latency. To address these limitations, we employ binary logarithms with regional error correction to approximate division operations. By leveraging approximation errors at boundary regions to formulate logarithm and antilogarithm offsets, our approach effectively reduces hardware complexity while minimizing the inherent errors of binary logarithm-based division. Additionally, we propose a six-stage pipelined hardware architecture, synthesized and validated on a Zynq UltraScale+ FPGA platform. The implementation results demonstrate that the proposed divider outperforms conventional division methods in terms of resource utilization and power savings. Furthermore, its application in image dehazing and object detection highlights its potential for real-time, high-performance computing systems. Full article

This work proposes a vision system mounted on the head of an omnidirectional robot to track pineapples and maintain them at the center of its field of view. The robot head is equipped with a pan–tilt unit that facilitates dynamic adjustments. The system architecture, implemented in Robot Operating System 2 (ROS2), performs the following tasks: it captures images from a webcam embedded in the robot head, segments the object of interest based on color, and computes its centroid. If the centroid deviates from the center of the image plane, a proportional–integral–derivative (PID) controller adjusts the pan–tilt unit to reposition the object at the center, enabling continuous tracking. A multivariate Gaussian function is employed to segment objects with complex color patterns, such as the body of a pineapple. The parameters of both the PID controller and the multivariate Gaussian filter are optimized using a genetic algorithm. The PID controller receives as input the (x, y) positions of the pan–tilt unit, obtained via an embedded board and MicroROS, and generates control signals for the servomotors that drive the pan–tilt mechanism. The experimental results demonstrate that the robot successfully tracks a moving pineapple. Additionally, the color segmentation filter can be further optimized to detect other textured fruits, such as soursop and melon. This research contributes to the advancement of smart agriculture, particularly for fruit crops with rough textures and complex color patterns. Full article

In the post-Moore era, the excessive amount of information brings great challenges to the performance of computing systems. To cope with these challenges, approximate computation has developed rapidly, which enhances the system performance with minor degradation in accuracy. In this paper, we investigate the utilization of an Artificial Intelligence of Things (AIoT) processor for approximate computing. Firstly, we employed neural architecture search (NAS) to acquire the neural network structure for approximate computation, which approximates the functions of FFT, DCT, FIR, and IIR. Subsequently, based on this structure, we quantized and trained a neural network implemented on the AI accelerator of the MAX78000 development board. To evaluate the performance, we implemented the same functions using the CMSIS-DSP library. The results demonstrate that the computational efficiency of the approximate computation on the AI accelerator is significantly higher compared to traditional DSP implementations. Therefore, the approximate computation based on AIoT devices can be effectively utilized in real-time applications. Full article

Accurate identification of rice diseases is crucial for improving rice yield and ensuring food security. In this study, we constructed an image dataset containing six classes of rice diseases captured under real field conditions to address challenges such as complex backgrounds, varying lighting, and symptom similarities. Based on the MobileViT-XXS architecture, we proposed an enhanced model named MobileViT-DAP, which integrates Channel Attention (CA), Efficient Channel Attention (ECA), and PoolFormer blocks to achieve precise classification of rice diseases. The experimental results demonstrated that the improved model achieved superior performance with 0.75 M Params and 0.23 G FLOPs, ensuring computational efficiency while maintaining high classification accuracy. On the testing set, the model achieved an accuracy of 99.61%, a precision of 99.64%, a recall of 99.59%, and a specificity of 99.92%. Compared to traditional lightweight models, MobileViT-DAP showed significant improvements in model complexity, computational efficiency, and classification performance, effectively balancing lightweight design with high accuracy. Furthermore, visualization analysis confirmed that the model’s decision-making process primarily relies on lesion-related features, enhancing its interpretability and reliability. This study provides a novel perspective for optimizing plant disease recognition tasks and contributes to improving plant protection strategies, offering a solution for accurate and efficient disease monitoring in agricultural applications. Full article

In complex agricultural settings, accurately and rapidly identifying the growth and health conditions of strawberries remains a formidable challenge. Therefore, this study aims to develop a deep framework, Disease-YOLO (D-YOLO), based on the YOLOv8s model to monitor the health status of strawberries. Key innovations include (1) replacing the original backbone with MobileNetv3 to optimize computational efficiency; (2) implementing a Bidirectional Feature Pyramid Network for enhanced multi-scale feature fusion; (3) integrating Contextual Transformer attention modules in the neck network to improve lesion localization; and (4) adopting weighted intersection over union loss to address class imbalance. Evaluated on our custom strawberry disease dataset containing 1301 annotated images across three fruit development stages and five plant health states, D-YOLO achieved 89.6% mAP on the train set and 90.5% mAP on the test set while reducing parameters by 72.0% and floating-point operations by 75.1% compared to baseline YOLOv8s. The framework’s balanced performance and computational efficiency surpass conventional models including Faster R-CNN, RetinaNet, YOLOv5s, YOLOv6s, and YOLOv8s in comparative trials. Cross-domain validation on a maize disease dataset demonstrated D-YOLO’s superior generalization with 94.5% mAP, outperforming YOLOv8 by 0.6%. The framework’s balanced performance (89.6% training mAP) and computational efficiency surpass conventional models, including Faster R-CNN, RetinaNet, YOLOv5s, YOLOv6s, and YOLOv8s, in comparative trials. This lightweight solution enables precise, real-time crop health monitoring. The proposed architectural improvements provide a practical paradigm for intelligent disease detection in precision agriculture. Full article

The evolution of smart cities is intrinsically linked to advancements in computing paradigms that support real-time data processing, intelligent decision-making, and efficient resource utilization. Edge and cloud computing have emerged as fundamental pillars that enable scalable, distributed, and latency-aware services in urban environments. Cloud computing provides extensive computational capabilities and centralized data storage, whereas edge computing ensures localized processing to mitigate network congestion and latency. This survey presents an in-depth analysis of the integration of edge and cloud computing in smart cities, highlighting architectural frameworks, enabling technologies, application domains, and key research challenges. The study examines resource allocation strategies, real-time analytics, and security considerations, emphasizing the synergies and trade-offs between cloud and edge computing paradigms. The present survey also notes future directions that address critical challenges, paving the way for sustainable and intelligent urban development. Full article

This study introduces a custom-designed CNN architecture that extracts robust, multi-level facial features and incorporates preprocessing techniques to correct or reduce asymmetry before classification. The innovative characteristics of this research lie in its integrated approach to overcoming facial asymmetry challenges and enhancing CNN-based emotion recognition. This is completed by well-known data augmentation strategies—using methods such as vertical flipping and shuffling—that generate symmetric variations in facial images, effectively balancing the dataset and improving recognition accuracy. Additionally, a Loss Weight parameter is used to fine-tune training, thereby optimizing performance across diverse and unbalanced emotion classes. Collectively, all these contribute to an efficient, real-time facial emotion recognition system that outperforms traditional CNN models and offers practical benefits for various applications while also addressing the inherent challenges of facial asymmetry in emotion detection. Our experimental results demonstrate superior performance compared to other CNN methods, marking a step forward in applications ranging from human–computer interaction to immersive technologies while also acknowledging privacy and ethical considerations. Full article

Neural networks (NNs) are essential in advancing modern safety-critical systems. Lightweight NN architectures are deployed on resource-constrained devices using hardware accelerators like Graphics Processing Units (GPUs) for fast responses. However, the latest semiconductor technologies may be affected by physical faults that can jeopardize the NN computations, making fault mitigation crucial for safety-critical domains. The recent studies propose software-based Hardening Techniques (HTs) to address these faults. However, the proposed fault countermeasures are evaluated through different hardware-agnostic error models neglecting the effort required for their implementation and different test benches. Comparing application-level HTs across different studies is challenging, leaving it unclear (i) their effectiveness against hardware-aware error models on any NN and (ii) which HTs provide the best trade-off between reliability enhancement and implementation cost. In this study, application-level HTs are evaluated homogeneously and independently by performing a study on the feasibility of implementation and a reliability assessment under two hardware-aware error models: (i) weight single bit-flips and (ii) neuron bit error rate. Our results indicate that not all HTs suit every NN architecture, and their effectiveness varies depending on the evaluated error model. Techniques based on the range restriction of activation function consistently outperform others, achieving up to 58.23% greater mitigation effectiveness while keeping the introduced overhead at inference time low while requiring a contained effort in their implementation. Full article

The integration of edge-to-cloud infrastructures in smart grid (SG) data center networks requires scalable, efficient, and secure architecture. Traditional server-based SG data center architectures face high computational loads and delays. To address this problem, a lightweight data center network (DCN) with low-cost, and fast-converging optimization is required. This paper introduces a container-based time synchronization model (CTSM) within a spine–leaf virtual private cloud (SL-VPC), deployed via AWS CloudFormation stack as a practical use case. The CTSM optimizes resource utilization, security, and traffic management while reducing computational overhead. The model was benchmarked against five DCN topologies—DCell, Mesh, Skywalk, Dahu, and Ficonn—using Mininet simulations and a software-defined CloudFormation stack on an Amazon EC2 HPC testbed under realistic SG traffic patterns. The results show that CTSM achieved near-100% reliability, with the highest received energy data (29.87%), lowest packetization delay (13.11%), and highest traffic availability (70.85%). Stateless container engines improved resource allocation, reducing administrative overhead and enhancing grid stability. Software-defined Network (SDN)-driven adaptive routing and load balancing further optimized performance under dynamic demand conditions. These findings position CTSM-SL-VPC as a secure, scalable, and efficient solution for next-generation smart grid automation. Full article