With the rapid advancement of 6G, the task offloading has emerged as a critical issue for enhancing computational efficiency in the Industrial Internet of Things (IIoT). However, industrial devices are often constrained by computing power, energy and mobility, challenging the delay-sensitive and compute-intensive tasks consisting of subtasks with complex dependencies, e.g., industrial image recognition. Given the increasing task complexity, developing efficient offloading strategies in dynamic multislot industrial scenarios with mobile devices remains a challenge. To address this issue, a task offloading scheme for industrial image recognition in digital twin edge networks (DITENs) is proposed. By leveraging the digital twin (DT) to accurately model the states of mobile industrial tasks and edge servers, the task offloading is formulated to optimize the weighted sum of task processing delay and energy consumption, that is proven to be NP-hard. Since the task of image recognition can be represented as a directed acyclic graph (DAG), the dependencies between subtasks are extracted using graph convolutional network (GCN) to generate optimal execution priorities for task offloading. Through proving the optimization problem as a Markov decision process (MDP), an improved multiagent deep deterministic policy gradient algorithm, named$\epsilon $-ATN-MADDPG, incorporating the$\epsilon $-greedy strategy and the self-attention mechanism to enable efficient decision-making in dynamic environments, is designed to offer a promising solution. Experimental results on the KolektorSDD dataset demonstrate that this solution outperforms compared methods.

In 6G-enabled vehicle-to-edge networks, through deploying computing power resources closer to mobile vehicles for providing low latency and highly reliable services, Mobile Edge Computing (MEC) as an emerging paradigm has promoting mobile vehicles with limited capacities to come with diversified artificial intelligence (AI) applications. Nevertheless, the computing power of MEC server is still finite, seeking the optimal resource pricing and allocation strategies for MEC servers, and determining the optimal computation offloading for intelligent vehicle applications, still remain challenging issues that are necessary to be reasonably solved to improve the service experience. To address this issue, a scenario that multiple intelligent vehicle applications cooperatively initiate computation offloading requests in 6G-enabled multi-server multi-access vehicle-to-edge computing systems is considered in this paper, and with the defined reasonable utilities for MEC servers and intelligent vehicle applications, a solution of bidding-enabled dynamic resource pricing for computation offloading is proposed. Concretely, through considering the relationship of resource supply and demand, and the bidding among MEC servers, a dynamic resource pricing scheme is designed for MEC servers, meanwhile, with the complete consideration of dynamic resource pricing and time-varying wireless channel interference in multi-cell networks, a Q-learning based offloading decision algorithm is proposed for intelligent vehicle applications. Simulation experiments finally are conducted to demonstrate the efficiency in achieving the win-win situation with guaranteed utilities for both MEC servers and intelligent vehicle applications.

As a typical deterministic network, the Industrial Internet of Things (IIoT) has strict requirements on the task processing delay. IIoT leverages edge computing to deliver low-latency services efficiently. However, this requires mobile Internet of Things (IoT) devices to execute reasonable task offloading decisions. To minimize disruption in IIoT operations and facilitate cost-effective execution, our approach capitalizes on the synergy of Digital Twin (DT) for enhanced virtual-real coordination in decision-making. We introduce an innovative solution for mobile inspection tasks within a DITEN-enhanced IIoT framework. This solution, grounded in A-MADDPG based partial offloading, not only addresses the challenges posed by IIoT but also transforms the problem into an optimization problem with constraints. By conceptualizing it as a Markov Decision Process (MDP), we integrate a carefully designed attention mechanism into the MADDPG algorithm, named as Attention enhanced Multi-Agent Deep Deterministic Policy Gradient(A-MADDPG), forging a robust and effective solution. Experimental results underscore the superiority of our proposed method, demonstrating remarkable convergence and outperforming existing baselines in reducing task processing latency.

Digital twin (DT) is a virtual representation of physical entities or processes that enables real-time monitoring, analysis, and optimization in the field of intelligent manufacturing. By simulating and optimizing production processes, DT technology could predict and prevent equipment failures, and enhance the efficiency and quality of industrial parts production. However, effectively deploying DTs into Digital Twin Empowered Edge Network (DITEN) in complex Industrial Internet of Things (IIoT) environments remains a significant challenge. Particularly in scenarios with numerous physical entities within IIoT, optimizing the deployment of DTs on edge nodes to minimize interaction latency with physical entities, as well as reducing workload and energy consumption on edge nodes, becomes crucial. To address this challenge, this paper first designs a Bi-Layer DT architecture for IIoT. Furthermore, an Intrinsic Curiosity Module-based Multi-Agent Proximal Policy Optimization algorithm (ICM-MAPPO) is proposed to solve the optimal deployment problem for DTs in DITEN-Empowered IIoT. Numerical experiments validate the effectiveness of the ICM-MAPPO algorithm in minimizing deployment latency and interaction latency while achieving load balance and reducing energy consumption.

Efficient image segmentation algorithms are critical in computer vision, as they maintain high processing speeds while handling large amounts of data and providing practical solutions in resource-limited environments. While existing classical segmentation methods achieve good results, their real-time performance can be further improved. To address this issue, we propose an efficient segmentation network (ESNet) to capture extensive contextual information and improve segmentation fineness, which are two main issues in semantic segmentation, while ensuring real-time capability. Then, we first introduce an efficient context module (ECM) that effectively enlarges the effective receptive field (ERF) of the model and improves its performance in integrating contextual information. Subsequently, a skip connection with a simple feature fusion module (SFFM) is designed to provide rich detail information, thereby improving segmentation fineness. The efficacy of ESNet was evaluated on the PASCAL VOC2012 semantic segmentation dataset against several state-of-the-art methods. ESNet achieves 71.4 FPS at a resolution of 1024×2048 on a single NVIDIA GeForce RTX 3090 GPU and achieves 62.42% mIoU at a resolution of 320×480 with an extremely simple training recipe, striking a fine balance between segmentation accuracy and inference speed.

The objective of smoke segmentation is to distinguish the smoke area from the background in smoky images, thereby providing essential information for the subsequent smoke removal process. However, current smoke segmentation algorithms consistently underperform in terms of accuracy. To address this challenge, we first developed a new smoke segmentation dataset comprising over 2000 images, each with comprehensive smoke annotations. The dataset can enhance the precision of smoke segmentation and serve as a foundation for training and testing. Then, we designed a smoke prior module to improve the efficacy of smoke removal. This module generates soft attention maps, which can strategically assign the weight of smoke details within the image. Finally, to validate the effect of smoke removal, we have also seamlessly incorporated the smoke prior module into the image fusion, which is crucial for enhancing the overall image clarity and detail. Extensive experimentation demonstrates that the proposed method outperforms state-of-the-art segmentation models in terms of segmentation accuracy and also achieves superior performance on smoke removal tasks.

The design and performance analysis of underwater acoustic sensor networks (UASNs) have attracted intensive attention. In this study, we present a theoretical framework to evaluate the performances of UASNs. In contrast to existing methods, we applied a new communication protocol model to characterize the channel interference by taking into account the unique characteristics of underwater acoustic channels, namely, the concurrent transmission opportunities owing to the large propagation delays of acoustic signals. Using the communication protocol model, we investigated the collision regions of a communication pair and then derived a closed-form expression for the successful packet transmission probability. Moreover, we considered practical networks where each node is equipped with a limited data buffer to store its generated packets. Using collision analysis and a stationary buffer distribution, we derived the closed-form expression of system performance in terms of the throughput and the packet queue delay. Finally, we validated the accuracy of our analysis via extensive simulation results. The impacts of various network parameters on the system performance are also evaluated.

Nowadays, video surveillance is widespread used to achieve security life in the construction of smart city. As a result, prevalence of video surveillance equipments and technologies enables pedestrian identification and tracking to be research hotspots in the field of computer vision, whose development and application are of great significance to the construction of a good social security environment. However, pedestrian identification and tracking in monitoring scenarios still have problems of low recognition accuracy and high model complexity, and computer vision based adaptive recognition methods still have a large room for improvement and development. To address this issue, real-time pedestrian identification and tracking within adaptive collaboration edge computing environment is investigated in this paper. Within the paradigm of edge computing, the Raspberry Pi 3B+ acting as the edge computing node is adopted to handle the related issues of pedestrian identification and tracking. The combination of HOG (Histogram of Oriented Gradient) and SVM (Support Vector Machine) is investigated to achieve pedestrian identification, where, HOG is employed as the feature descriptor and SVM is employed as the classification algorithm. Furthermore, the pixel-based visual tracking algorithm is investigated to achieve effective and uninterrupted pedestrian tracking. By implementing a prototype on Raspberry Pi 3B+ using OpenCV libraries, the experimental results finally have been shown to demonstrate the efficiency of the investigations.

Intelligent transportation is an important guarantee for the safety and efficiency of urban transportation in smart cities, and regular road pavement inspection is the focus of road and bridge maintenance in intelligent transportation. Cracks in concrete pavement are the most common type of pavement damage and the earliest sign of pavement deterioration. However, existing crack detection algorithms suffer from incomplete crack detection and are easily disturbed by pseudo-cracks such as water spots and leaves. To address the above problems, this paper proposes a convolutional neural network (CNN) method that introduces a gradient module and an attention mechanism. The method adopts a CNN model based on the VGG-16 structure as the main body of the network structure, and optimally adjusts the network structure by incorporating a gradient layer and a self-attention mechanism, accelerating the convergence speed of network training and the global information learning ability. A negative sample dataset with pseudo-cracks, such as leaves, water spots and branches was constructed, and comparative experimental analysis was conducted in terms of both visual judgment and objective indicators. The experimental results show that after the introduction of the gradient layer and the self-attentive mechanism, not only the convergence speed of the network training is faster, but also the cracks in the concrete pavement images can be segmented more completely and accurately.

With the development and application of Internet of Things (IoT) technology, IoT has been widely used in agriculture, industry, and urban construction. In the process of building Smart cities, setting up small-scale weather stations based on IoT technology can effectively monitor certain special environments where large weather stations cannot accurately assess and predict short-term extreme weather events. As the summer heat approaches, the number of heatstroke cases is continuously rising. The Wet Bulb Globe Temperature (WBGT) index, which is closely related to heatstroke, provides a simple method for evaluating the thermal work environment and thermal load of workers in hot conditions. In this paper, through adjusting the model structure and moving window, and optimizing the predictive model parameters, an improved Long Short Term Memory (LSTM) algorithm is proposed to forecast WBGT values at future time points: five minutes, thirty minutes, and sixty minutes ahead. The paper also conducts a simple analysis of the daily average performance of WBGT in relation to air humidity. Additionally, it compares dual-input models that include air humidity as input. Through a comparison of four prediction performance evaluation metrics, simple-input LSTM models demonstrate lower error.

In this letter, we propose a novel algorithm to detect the optimal seamline by using the edge-enhanced energy function for orthoimage mosaicking. First, we generate the energy cost map by using traditional intensity and gradient difference. Then, to ensure that the detected optimal seamlines avoid crossing the obvious objects, the energy map is enhanced in the regions of valid edge segments around the obvious objects. The edge segments are separately detected from input images, and the invalid edge segments are removed by using texture complexity index, image difference, and consistency constraint. Finally, we detect the optimal seamline from this edge-enhanced energy cost map via graph cuts. Experimental results on several groups of challenging orthoimages show that the proposed algorithm is capable of creating high-quality seamlines for orthoimage mosaicking and outperforms state-of-the-art algorithms and the commercial software based on the visual comparison and statistical evaluation.