Search

In this era of extensive big-data, efficient processing of this data has become a critical aspect of computing. Resistive random access memory devices have exemplary non-volatile memory properties and have received significant research attention as next-generation devices for emulating neuromorphic computing. In this study, we present a synaptic memristor incorporating a thin Al-doped hafnium oxide (HfAlO) film layer sandwiched between two electrodes, which is designed to emulate various functions of the biological brain. Investigations into endurance and retention revealed the non-volatile nature of the memristor, with consistent resistive switching observed during continuous operation and a low operating voltage requirement of less than 1.7 V. Additionally, by applying different conditional stimuli, we successfully implemented Pavlovian associative learning. Furthermore, using sequential pulses with varying sequences resulted in the creation of 4-bit edge computing, demonstrating capabilities for energy- and time-efficient data processing. The synaptic and computing properties exhibited by the Mo/HfAlO/TiN device highlighted its valuable features, positioning the device as a promising candidate for energy-efficient neuromorphic computing hardware in the field of artificial intelligence. • CMOS compatible memristor is designed with material analysis. • Highly controllable memristive switching is achieved in HAO memristor. • Artificial synapse behaviors are emulated by pulse modulation. • System level neurormorphic simulation evaluates the performance of each operation. • Edge-computing function is conducted. [ABSTRACT FROM AUTHOR]

• Multi-modal fusing algorithm with rice disease image and non-imaging spectral signals. • Image feature extraction with a new neural network root in MobilenetV2. • Linear SVM method with fused features obtained by effective fusion strategy for crop disease classification. Frequent occurrences of pests and diseases in farmland pose a severe threat to crop yield and quality. Various techniques utilized for the automatic detection of diseases and pests are mostly based on a single modality with limited diagnostic information. In this study, a multi-modal classification algorithm, ISMSFuse, was proposed to fuse the image and spectral information. This algorithm combines a MobilenNetV2 lightweight neural network with a traditional low-computational linear SVM method to adapt to edge computing terminals. The spectral information was extracted with non-imaging spectral signals to reduce the acquisition costs and hardware requirements. A proprietary multi-mode rice bacterial blight dataset, denoted as ImgS-RBB2022, was curated specifically for evaluating the algorithm. The effectiveness of the image features extracted by the algorithm model was evaluated by the t-SNE method and the Fisher scores. Finally, comparing the results with only using imaging data, the multi-mode fusion algorithm had a superior recognition accuracy of 98.14%. Through the SHAP method, it was shown that the added spectral features became the most critical feature affecting the output of the algorithm model. To further explore the application potential of the proposed algorithm, we deployed it on a board-based microcomputer Raspberry Pi to test its feasibility in an edge computing circumstance. The proposed method can be applied to various agricultural diseases and pest detection tasks, particularly in the case of foliar-type diseases and pests. [ABSTRACT FROM AUTHOR]

The ever-increasing number of recording sites of silicon-based probes imposes a great challenge for detecting and evaluating single-unit activities in an accurate and efficient manner. Currently separate solutions are available for high precision offline evaluation and separate solutions for embedded systems where computational resources are more limited. We propose a deep learning-based spike sorting system, that utilizes both unsupervised and supervised paradigms to learn a general feature embedding space and detect neural activity in raw data as well as predict the feature vectors for sorting. The unsupervised component uses contrastive learning to extract features from individual waveforms, while the supervised component is based on the MobileNetV2 architecture. One of the key advantages of our system is that it can be trained on multiple, diverse datasets simultaneously, resulting in greater generalizability than previous deep learning-based models. We demonstrate that the proposed model does not only reaches the accuracy of current state-of-art offline spike sorting methods but has the unique potential to run on edge Tensor Processing Units (TPUs), specialized chips designed for artificial intelligence and edge computing. We compare our model performance with state of art solutions on paired datasets as well as on hybrid recordings as well. The herein demonstrated system paves the way to the integration of deep learning-based spike sorting algorithms into wearable electronic devices, which will be a crucial element of high-end brain–computer interfaces. • Deep learning methods can generalize on spike sorting data. • Contrastive learning delivers high-quality feature space for spike waveforms. • MobileNet architecture detects spikes at a state-of-art performance. • Edge TPU can be used to perform fast spike sorting on devices with limited resources. [ABSTRACT FROM AUTHOR]

With the advancement in the Internet of Things (IoT) and Wireless Sensor Node (WSN) applications computation on edge is preferred over computation on the cloud. However, due to limited resource availability in the edge environment, energy efficiency can be seen as the primary barrier in the growth of edge computing devices. In edge computing applications like wearable devices, forest monitoring systems, and other such applications, power is of utmost importance. The use of a complex, pipelined, general-purpose processor is redundant for such low-end applications and it leads to high power consumption. Hence, edge computing systems require low-power solutions while still maintaining a minimum performance level. In this work, an energy-efficient single-cycle RISC-V instruction set architecture (ISA) based RV32I processor is designed for low-end edge computing applications. The design incorporates base integer implementation of RISC-V ISA. The proposed computing system is designed using Verilog HDL and synthesized on Semi-conductor Laboratory (SCL) 180 nm CMOS process technology node. The synthesis tool reports power consumption of 14.7 mW at 50 MHz frequency with an area of 0.24 mm2. Though synthesis of this design reports area of 0.24 mm2, the design is taped out as a part of an available shuttle project where die size is fixed to 5 mm × 5 mm. The PNR (Place and Route) tool reports power consumption of 11.8 mW at an operating frequency of 40 MHz. A basic functional testing of proposed fabricated RV32I chip is carried out in laboratory experimental setup and results are presented. The comparison result shows that the proposed RV32I processor consumes 0.29 mW/MHz power which is less than state-of-the-art low-end processors. Hence, the proposed design is suitable for low-end edge computing applications. • For low-end edge computing applications, the use of a complex, pipelined, general-purpose processor is redundant, and it leads to high power consumption. • In this work, an energy-efficient single-cycle RISC-V ISA-based RV32I processor is designed for low-end edge computing applications. • The proposed processor architecture is implemented using Verilog HDL and prototyped on FPGA. • Further, the proposed design is tapedout for ASIC implementation on 180 nm CMOS process technology node. • The fabricated RV32I chip is validated in laboratory experimental setup. [ABSTRACT FROM AUTHOR]

In recent years, the rapid development of UAV technology has greatly improved the efficiency of the detection of concrete bridge cracks. With the increase in the number of bridge inspection UAVs, the number of tasks handled by cloud services has increased linearly, resulting in increased computational pressure on cloud services. In order to reduce the computational load of cloud servers, we proposed a crack segmentation network based on UAV-enabled edge computing. However, due to the limitation of computational capability of edge computing and the strength inhomogeneity and background complexity of cracks, crack detection is still a challenging task. Thus, we proposed an effective concrete crack segmentation network based on UAV-enabled edge computing, the network used feature map fusion to fuse different levels of feature map information into lower-level features for crack detection. The atrous spatial pyramid pooling network was used to increase the low-resolution feature map receptive field information for cracks and to enhance the detection accuracy for cracks of different scales. In addition, loss functions for crack datasets were proposed to solve the problem of imbalance due to positive and negative samples in the concrete crack images. Experiments demonstrated that the proposed methods are better than the state-of-the-art edge detection and semantic segmentation methods in terms of accuracy and generality. [ABSTRACT FROM AUTHOR]

Edge computing is an emerging computing paradigm, which offers a great opportunity to implement data mining-based services and applications for a large number of devices and sensors in Internet of Things. However, the new paradigm is faced with security and privacy challenges due to the diversity and the limited capability of edge components. In particular, data privacy is one of the most concerned problems for all the participants. In this paper, we propose a framework of privacy-preserving data mining based on private random decision trees in edge computing, which not only gives the strong privacy guarantee, but also provides a certain amount of data utility. Firstly, we design a preservation framework to implement private random decision trees satisfying local differential privacy. Secondly, we present the concrete implementations of algorithms and the corresponding task that each participant needs to undertake. Thirdly, we analyze the key factors to influence privacy and utility, including the allocation of data and privacy budget. Fourthly, we give the improved algorithms to further increase the utility with strong privacy preservation. Finally, extensive experiments demonstrate the good performance of our designed framework. [ABSTRACT FROM AUTHOR]

Modern sensor technologies increasingly enrich studies in wildlife behavior and ecology. However, constraints on weight, connectivity, energy and memory availability limit their implementation. With the advent of edge computing, there is increasing potential to mitigate these constraints, and drive major advancements in wildlife studies. [ABSTRACT FROM AUTHOR]

With the rapid development of cloud computing, Internet of things and artificial intelligence, human–computer interaction (HCI) is playing an increasingly important role in the daily life. As an important component of HCI, hand gesture recognition (HGR) system is usually combined with edge computing server, utilizing machine learning, including neural network, decision tree, integrated learning, to achieve low latency and high reliability service. High precision HGR with low computational complexity is prerequisite for the commercialization of gesture recognition. Therefore, this paper proposed a high-precision parameter correction algorithm based on the established scattered-point model and the outlier detection scheme, and a recognition algorithm with multivariable decision tree is then presented for the dynamic hand gestures. The experimental results show that the proposed algorithms can improve the recognition accuracy and effectively reduce the running time, which is conducive to algorithm transplantation and model deployment in edge servers. [ABSTRACT FROM AUTHOR]

The computation offloading problem in mobile edge computing (MEC) has received a lot of attention, but service caching is also a research topic that cannot be ignored in MEC. Due to the limited resources available on the Edge Server (ES), a wise computation offloading and service caching policy must be formulated in order to maximize system offload efficiency. In this paper, a many-objective joint optimization computation offloading and service caching model (MaJOCOSC) is designed. The model takes into account the limited computing and storage resources of ES, the delay and energy consumption constraints of different types of tasks, and multiple processing modes of user tasks, and sets delay, energy consumption, task hit service rate, service cache balancing, and load balancing as the five optimization objectives of MaJOCOSC. Meanwhile, a non-dominated sorting genetic algorithm (NSGAIII-ASF&WD) based on achievement scalar function (ASF) and the k-nearest neighbor weighted distance mating selection strategy is proposed for better solving the model. The ASF ensures that the given strategy performs well for each objective value, and the k-nearest neighbor weighted distance provides the user with a diversity of strategies. Simulation results show that NSGAIII-ASF&WD can obtain better objective values when solving the model compared with other many-objective evolutionary algorithms, and a suitable computation offloading and service caching strategy is obtained. [ABSTRACT FROM AUTHOR]

• A new in-situ monitoring system was proposed to upscale resolution and extract geometric characteristics during AM process. • The proposed ViTSR performs better than the SOTA methods in terms of PSNR, which may reduce the cost of high-speed camera. • The molten pool and plasma arc can be simultaneously and accurately extracted by FCN with input of reconstructed frames. • The low-cost AI edge board reduces the inference time of ViTSR to 50.97 ms, which reduce the cost of computing in AM process. In additive manufacturing (AM), in-situ monitoring systems are vital for ensuring process quality. However, the widespread adoption of traditional high-speed camera-based monitoring systems is hindered by the prohibitively high cost of the required high-speed cameras. This paper introduces an innovative low-cost in-situ monitoring system that utilizes AI edge computing boards to expedite digital image processing without requiring high resolution (HR) video sequences. The system integrates a visual transformer-based video super resolution (ViTSR) network for reconstructing high resolution video frames and employs a fully convolutional network (FCN) to extract geometric characteristics of the molten pool and plasma arc simultaneously during AM processes. Comparing ViTSR with six state-of-the-art super-resolution methods, it achieved the highest peak signal-to-noise ratio (PSNR) of 38.16 dB on the test data. Moreover, the FCN utilized the reconstruction results from ViTSR, demonstrating an accuracy of 96.34% in multi-object extraction tasks. Through operator fusion and model pruning, the inference time of ViTSR and FCN on the AI edge board is optimized to 50.97 ms and 67.86 ms, respectively. Consequently, the proposed system achieves a total inference time of 118.83 ms, meeting the real-time quality monitoring needs of AM processes. Furthermore, this approach effectively reduces the expenses associated with high-speed camera-based monitoring systems, thereby promoting the widespread adoption of in-situ monitoring systems within the realm of AM. [ABSTRACT FROM AUTHOR]

Big data will cause the system for business public opinion to be overburdened, and dynamic changes in computing resources will cause tasks to be delayed. To end the situation, this study assigns non-fixed execution time to each task node based on the task soft deadline and task constraints and solves the problem of difficult task scheduling caused by task dependency constraints. Aiming at the problem of task delay caused by dynamic changes of computing resources, this paper proposes an adaptive offloading and scheduling algorithm for dependent tasks in the mobile edge computing environment. This study takes the economic efficiency analysis system as an example and uses the resource allocation management algorithm based on linked lists and edge servers to study the communication resource allocation management of the economic efficiency analysis system. In addition, this study designs experiments to perform performance analysis of the algorithm proposed by this study. The research results show that the proposed algorithm has an obvious effect. [ABSTRACT FROM AUTHOR]

In the realm of multi-UAV-assisted mobile edge computing (MEC), the joint optimization of UAV deployment and task scheduling has emerged as a potent approach for mitigating energy consumption while ensuring high-quality services for mobile users. Nevertheless, the intricate interdependence between UAV deployment and task scheduling presents formidable challenges when implementing this scheme. In response, we introduce a novel methodology called the Dual-Stage Hybrid Strategy with Genetic-Simulated Annealing and Knowledge Sharing (DSHSGSK). DSHSGSK employs an alternating optimization strategy known as HS-GSK, supplemented by an elimination operator, to simultaneously optimize UAV deployment and task scheduling. In pursuit of heightened efficiency in UAV deployment and a more effective task scheduling strategy, we propose an innovative location encoding scheme for UAVs. To harness the local search capabilities of the Harmony search (HS) algorithm and the global search capabilities of the Gaining Sharing Knowledge based Algorithm (GSK), we introduce a coordination parameter (CP) to modulate the frequency of alternation between these two optimization techniques. Experimental results validate the superiority of our proposed method over other state-of-the-art algorithms for jointly optimizing multi-UAV deployment and task scheduling. Notably, our approach excels in terms of energy consumption, underscoring its exceptional performance. [ABSTRACT FROM AUTHOR]

Internet of Vehicles (IoV) relies heavily on its computing capability to facilitate various vehicular applications. Since the cloud computing or mobile edge computing (MEC) only cannot satisfy the latency requirement due to the limitation of the resource coverage, the cloud–edge-end cooperative computing has become an emerging paradigm. A comprehensive IoV architecture is considered and a joint optimization problem is formulated to minimize the system function value. To optimize the resource allocation and the task offloading strategy, the simulated spring system algorithm (SSSA) is designed where the initial problem is decoupled into two sub-problems with priority. The first one is to allocate computing resources based on KKT conditions, thus the individual optimal solution is achieved. The second one is solved based on the idea of simulated spring system such that the task offloading strategy is obtained. Two sub-problems iterate mutually to update each other until finishing the binary tree traversal. Thus, the proposed solution adapts to various conditions and the computational complexity is also reduced compared with traditional methods. Simulation verifies that the proposed algorithm reduces the maximum system function value by about 31% compared with the benchmark methods and performs efficiently in various road conditions. [ABSTRACT FROM AUTHOR]

The rapid expansion of Mobile Edge Computing (MEC) is driven by the growing demand for resource-intensive applications within the Internet of Things. Computation offloading allows these applications to be executed at the network edge, but it leads to significant electricity expenses for network operators. To mitigate these costs, one promising solution is to power base stations (BSs) with a hybrid energy supply that combines unpredictable harvested energy with stable energy from the smart grid. This paper investigates joint computation scheduling for mobile devices (MDs) and resource allocation in a MEC network incorporating hybrid energy sources. Our objective is to maximize long-term time-averaged service utility by optimizing parameters such as BS battery supply, harvestable energy, CPU frequency, transmission power, task-partition factor, and MD-BS associations. To tackle this complex problem, we exploit the Lyapunov optimization framework to decompose it into deterministic subproblems for each time slot and propose an online network service utility maximization scheduling (NSUMS) algorithm. Experimental results show that our algorithm outperforms benchmark schemes in service utility and energy expenditure, improving the completion ratio by 32%, reducing the failure rate by 80%, and decreasing MD energy consumption by 28%. [ABSTRACT FROM AUTHOR]

As a branch of cloud computing, edge computing has received extensive attention. Edge computing offloads computing-intensive tasks to the computing-capable devices (i.e., edge servers) at the network edge of the cloud, and the edge servers assist in mobile devices to perform data analysis, which can significantly reduce the execution time of computing-intensive tasks on mobile devices. However, due to the poor quality of wireless channel communication, it may degrade the overall quality of the edge offloading when the mobile devices are linked to edge servers, that is, the communication link between the mobile device and the edge server may become a bottleneck. In this study, we propose a new edge computing scheme to support remote object detection by splitting the deep neural network (DNN), and we design a framework to reduce the amount of data transmitted over wireless links. Specially, the DNN is divided into a head modular and a tail modular, which are deployed on mobile devices and edge servers, respectively. Furthermore, the wireless link transmits the output of the last layer in the head modular to the edge server, instead of sending all the initial tensors of images to the edge server. And we employ three object detection models with complex structures, that is, Faster RCNN (regions with CNN features), Mask RCNN and Keypoint RCNN as the basic models, and modify their network architectures from the following aspects: 1) we introduce the information bottleneck principle and split the network architecture into a head modular and a tail modular, and we perform knowledge distillation to compress the head modular; 2) we embed a neural filter (NF) into the head modular to pre-filter images with no detection objects by using a simple convolutional neural (CNN) network, and the NF performs a binary classification task in the head modular. Lastly, we conduct experiments on the real COCO 2017 dataset and the PASCAL VOC 2012 dataset, and the results show that the proposed method outperforms other baselines. Especially, in person keypoint detection, the mAP (mean average precision) of the proposed method is at least 1.64% higher than other baselines on the COCO 2017 dataset. In addition, the prediction latency of the proposed splitting scheme at the data transmission rate of 1Mbps is at least 49.17% lower than that of local computing scheme. [ABSTRACT FROM AUTHOR]

Approximate computing has emerged as an efficient design methodology for improving the performance and power-efficiency of digital systems by allowing a negligible loss in the output accuracy. Dedicated hardware accelerators built using approximate circuits can solve power-performance trade-off in the computationally complex applications like deep learning. This paper proposes an approximate radix-4 Booth multiplier and hardware accelerator for deploying deep learning applications on power-restricted mobile/edge computing devices. The proposed accelerator uses approximate multiplier based parallel processing elements to accelerate the workloads. The proposed accelerator is tested with matrix–vector multiplication (MVM) and matrix–matrix multiplication (MMM) workloads on Zynq ZCU102 evaluation board. The experimental results show that the average power consumption of the proposed accelerator reduces by 34% and 40% for MVM and MMM respectively, as compared to the conventional multiply-accumulate unit that was used in the literature to implement similar workloads. Moreover, the proposed accelerator achieved an average performance of 5 GOP/s and 42.5 GOP/s for MVM and MMM respectively at 275 MHz, which are 14 × and 5 × respective improvements over the conventional design. • Low power and high speed approximate radix-4 Booth multiplier. • Approximate hardware accelerator for edge computing applications. • Simultaneous transfer of data from the on-chip memory to off-chip memory using multiple interconnects. • Packing of data for improving data communication bandwidth. [ABSTRACT FROM AUTHOR]

In cloud storage, attribute-based keyword searchable encryption realizes multi-user fine-grained authorization access control and outsourcing data sharing. However, (1) resisting key regeneration is an urgent problem to be solved in practical applications of attribute-based keyword search; (2) the general attribute-based keyword search scheme ignores the application scenario where multiple data owners jointly manage files. In addition, the interaction of data between intelligent terminals and cloud servers has become more frequent, leading to issues such as central load, transmission efficiency, and data privacy protection in cloud computing. To address the above issues, we propose an attribute-based searchable encryption scheme with the multi-data owner based on edge computing. In this scheme, the Pedersen (k , n) secret sharing protocol and the Reciprocal protocol are used to realize the common management and sharing of data by the multi-data owner to enhance data security. Secondly, the attribute set and identity information of each user are embedded into the private key to prevent key regeneration and resist user collusion attacks. Thirdly, edge computing is introduced, and part of the storage and decryption work is outsourced to edge nodes to reduce the computing load of users and data transmission bandwidth. Security proof and performance analysis show that the proposed scheme has strong security and practicability in multi-user sharing scenarios such as Online Education with multi-data owners. [ABSTRACT FROM AUTHOR]

In recent years, hardware sorters have been an attracted topic for researchers. Since hardware sorters play a crucial role in embedded systems, several attempts have been made to efficiently design and implement these sorters. Previous state-of-the-art hardware sorters are not suitable for embedded edge computing devices because they (1) consume high power, (2) occupy high area, (3) work for limited data-width numbers, (4) require many memory resources, and (5) finally, their architecture is not scalable with the number of input records. This paper proposes a hardware sorter for edge devices with limited hardware resources. The proposed hardware sorter, called Edge-Sorter, processes 4 bits of input records at each clock cycle. Edge-Sorter utilizes the unary processing in its main processing core. Edge-Sorter has valuable attributes compared to previous state-of-the-art techniques, including low power consumption, low area occupation, sorting numbers without storing their indices, sorting numbers with arbitrary data-width, and scalable with the number of input records. The proposed approach is evaluated and compared with previous state-of-the-art techniques with two different implementation and synthesis environments: Xilinx Vivado FPGA-based and Synopsys Design Compiler 45-nm ASIC-based. The Synthesis results of both environments indicate that both Edge-Sorter techniques reduces area and power consumption on average by 80% and 90%, respectively compared to previous techniques. [ABSTRACT FROM AUTHOR]

Edge computing, smart sensors, and health monitoring are boosting current wearable development and enabling the next technological user-centred revolution. Within this context, high added-value applications based on physiological information are gaining attention during the last years. Among the vast physiological metrics available, heart rate variability (HRV) is one of the most used. From such metric, different types of information related to the activity of the autonomic nervous system can be obtained. This fact has led integrated chip manufacturers to foster the design of novel analog front end circuitry for heart rate monitoring, which boosted a wearable smart sensor innovation. Notwithstanding the capabilities and efficiency of these novel sensors, different design space exploration (DSE) procedures must be addressed for every sensor integrated within any wearable system towards maximising the embedded resource usage. On this basis, this paper presents a DSE breakup for every stage involved in a wearable edge device developed by the authors and based on continuous HRV physiological monitoring. The particularities of such a system are detailed and explained. Moreover, time complexity and memory usage comparisons regarding different digital signal processing techniques are provided, which results in a set of potential recommendations for wearable constrained application needs. Finally, a use case is presented based on a rapid stress detection application by using the different DSE recommendations for our specific wearable edge device. This application reaches adequate trade-off precision for detecting physiological HRV activation using only a four-second temporal processing window. • Design Space Exploration is an essential task when facing the development of physiological edge computing devices. • The different morphology of PPG signals directly affects the data processing chain of HR-based wearable devices. • Heart Rate Variability is a reliable metric to detect physiological activation. [ABSTRACT FROM AUTHOR]

The proliferation of wireless applications at an exponential rate has made spectrum problems worse. Saturation in the unlicensed frequency spectrum is rapidly increasing as a result of the increasing data rates required by new wireless devices. A proposed solution to this problem is cognitive radio, which allows for the opportunistic use of licenced spectrum in less crowded areas. Cognitive network-based security evaluations using mobile edge computing and a Beyond 5G' (B5G) machine learning (ML) model are the focus of this research. In this case, the security study was carried out using cognitive network data transfer and multi-agent reinforcement encoder neural network and mobile edge computing (MRENN-MEC), a multi-agent reinforcement encoder neural network with mobile edge computing. Scalability, quality of service, throughput, and forecast accuracy are some of the network properties that undergo experimental analysis. [ABSTRACT FROM AUTHOR]

Unmanned aerial vehicle (UAV) and building information modeling (BIM) have been applied to inspect building surfaces to improve efficiency and reduce labor costs. However, most of the current research focus on the inspection mode of "first collect and then identify" with limited intelligence. Therefore, this study proposes a "global-local" adaptive inspection method for building surfaces by integrating BIM, UAV (with multifunction camera), and edge computing, enabling real-time synchronization of crack data collection and analysis. First, the building surface area to be inspected is extracted from BIM to generate global inspection points using the selected camera's field of view (FOV) and considering coordinates transformation between BIM and reality. Then, A star algorithm integrating with genetic algorithm (GA) is used to optimize the global flight path based on generated inspection points. Finally, the coordinates of the inspection points based on BIM are converted into real-world coordinates, and the flight parameters such as the flight yaw angle (FYA) and the gimbal yaw angle (GYA) are calculated to generate the inspection mission. During global inspection, a local adaptive inspection is proposed for each FOV through real-time local crack identification and flight replanning using edge computing and zoom camera. In addition, multiple battery supply points are automatically added considering the endurance of the UAV. A real building has been used to validate the proposed "global-local" adaptive inspection method. The experiment results show that the inspection time has been significantly reduced compared with the traditional method, and the crack identification of building surfaces can be implemented with high efficiency, precision, quality, and low cost. • A G-L adaptive inspection method is proposed to realize real-time synchronization of crack data acquisition and analysis. • A simpler BIM-based approach is developed for inspection path planning and optimization using A * algorithm and genetic algorithm. • Edge computing and multi-function camera are used to enhance the real-time environmental perception ability of the UAV. • The endurance of UAV has been fully considered, and multiple BSP paths can be automatically planned. [ABSTRACT FROM AUTHOR]

• Proposing a compact detection framework targeting online, real-time action detection. • Enhancing inference efficiency by feature approximation. • Detection accuracy is improved by adaptive multi-frame feature aggregation over time. • Achieving 70.92 and 49.53 F-mAP on public benchmark datasets UCF-24 and JHMDB-21 at 75 FPS. • Demonstrating an alternative detection solution for resource-constrained edge devices. Interpreting human actions requires understanding the spatial and temporal context of the scenes. State-of-the-art action detectors based on Convolutional Neural Network (CNN) have demonstrated remarkable results by adopting two-stream or 3D CNN architectures. However, these methods typically operate in a non-real-time, ofline fashion due to system complexity to reason spatio-temporal information. Consequently, their high computational cost is not compliant with emerging real-world scenarios such as service robots or public surveillance where detection needs to take place at resource-limited edge devices. In this paper, we propose ACDnet, a compact action detection network targeting real-time edge computing which addresses both efficiency and accuracy. It intelligently exploits the temporal coherence between successive video frames to approximate their CNN features rather than naively extracting them. It also integrates memory feature aggregation from past video frames to enhance current detection stability, implicitly modeling long temporal cues over time. Experiments conducted on the public benchmark datasets UCF-24 and JHMDB-21 demonstrate that ACDnet, when integrated with the SSD detector, can robustly achieve detection well above real-time (75 FPS). At the same time, it retains reasonable accuracy (70.92 and 49.53 frame mAP) compared to other top-performing methods using far heavier configurations. Codes will be available at https://github.com/dginhac/ACDnet. [ABSTRACT FROM AUTHOR]

The advancement of technology towards the realization of the evolving mobile computing paradigm brings a rapid paradigm shift in its usage, especially in the Internet, computation, and communications, that has a profound impact on businesses, services, and users. With the rise in resource-intensive or edge-based mobile applications such as autonomous driving, Amazon Go, virtual and augmented reality, and healthcare-related applications, countless challenges in computation and communication parameters like latency, bandwidth, and energy consumption are evolving. As a result, the multi-access edge computing (MEC) paradigm receives enormous attention where some portions of the user applications are offloaded to powerful machines for their efficient execution to optimize different evaluation metrics or to achieve performance goals. While a few survey works are available in this direction, none of them focuses explicitly on the emerging machine learning (ML) based computation offloading techniques and various associated sub-problems together. This paper aims to provide a detailed but precise overview of the research on using ML techniques for MEC environments. In this survey, we focus on how authors and researchers utilize the ML models in computation offloading problems on MEC architecture. We extend our study by considering several edge architectures, offloading parameters, ML approaches, and problem formulation strategies concerning computation offloading. Additionally, this paper discusses the potential challenges in the direction of computation offloading on MEC architecture. [ABSTRACT FROM AUTHOR]

Recent intelligent diagnostic algorithms for industrial practice have achieved impressive results. However, due to safety considerations, complex environments and deployment cost constraints in real industrial production, there are still some challenges in the form of edge computing as a vehicle. Therefore, this work proposes a fault diagnosis and distance localization model and deploys the model on edge computing devices for real-time fault diagnosis and distance localization. Lightweight is achieved by modifying the embedding module of the Transformer module as well as self-attention (MFLOPs: 0.487, Params: 6.148 k). Simultaneously, the introduction of D dformable C o n colution in self-attention and multidimensional output in decoding achieves strong robustness (95.92% of accuracy and ± 1.32 cm of error in the noise environments with SNR of −10 dB to 10 dB). Finally, this method was deployed on two edge computing devices for real-time diagnosis (FPS: 34 and 38). As the first attempt to explore the feasibility of deploying fault diagnosis and distance location strategies in edge computing devices, this work not only provides a novel solution for the challenge in industrial practice, but also is expected to promote the process of advanced information processing technology from theoretical design to industrial practice. [ABSTRACT FROM AUTHOR]

Vehicular Edge Computing (VEC) provides flexible distributed computing paradigm for vehicular network through computation offloading. With the advent of a growing of modern vehicle applications, the challenge for VEC network to fulfill the expansive demands from various suppliers, users and environments is increasingly prominent. This paper formulates a many-objective computation offloading problem in heterogeneous VEC network aiming to fulfill the diversified optimization requirements, including minimizing the task completion time, energy consumption and resource costs as well as load balance. A many-objective optimization algorithm named MaOITGO-CO is proposed to solve the formulated problem based on ITGO (Invasive Tumor Growth Optimization) by simulating the growth patterns of tumor cells. Specifically, considering the characteristics of computation offloading in VEC scenarios such as mobility, real-time requirements and the variety of tasks and resources, four types of tumor cells are equipped with different search strategies to enhance the search effectiveness and efficiency. The simulation results show that the proposed approach can provide high quality Pareto solutions for computation offloading problem, which outperforms other widely used algorithms in terms of convergency and diversity. Furthermore, the results of scalability experiments validate the availability of MaOITGO-CO when the problem is extended to different scales of both tasks and computing resources. [ABSTRACT FROM AUTHOR]

Vehicular Edge Computing (VEC) is a promising paradigm for providing low-latency and high-reliability services in the Internet of Vehicles (IoV). The increasing number of mobile devices and the diverse resource requirements of the growing IoV have resulted in a shift from centralized resource management to a decentralized approach. This shift offers improved fault tolerance, scalability, and privacy preservation. However, constructing collaborative awareness and coordination mechanisms between multiple vehicles and edge nodes in a decentralized manner is a challenge. To address this issue, we propose a decentralized many-to-many task offloading method that aims to minimize the average task completion latency of vehicles. In this study, we propose a data-sharing mechanism between vehicles and edge servers using the digital twin service, which provides global environmental perceptions to the vehicles by a low-cost approach. Additionally, we develop a mean-field multi-agent reinforcement learning algorithm to generate coordinated task offloading schemes. Instead of interacting with multiple agents, the vehicle only needs to respond to the mean action of the environment. Based on this transition, the agent generates coordinated task offloading decisions in complex scenarios. We evaluate the performance of our method using real urban traffic data. Experiment results verify the efficiency of our proposed method. [ABSTRACT FROM AUTHOR]

Mobile edge computing is an emerging technology paradigm that improves mobile devices' computational power and efficiency by offloading computational tasks to edge servers. However, there are still some problems: (1) Ignoring the complex time dependencies of compute-intensive or latency-sensitive tasks, leading to the omission of offload scheduling opportunities for fine-grained sub-tasks within a task, thus reducing the computational power of the mobile device and the utilization of edge resources; (2) Overly biased towards optimizing task execution latency and ignoring the balance between data transmission latency and energy loss, reducing the scheme versatility in natural computing environments. Aiming at the above problems, this paper models the complex dependencies, transmission time, and execution time between computing tasks, constructs a mathematical model of the scheduling and offloading process of computing tasks, and a joint task scheduling and offloading method (N-JSU) based on Nash equilibrium is designed. This offloading strategy is divided into two steps: first, the average latency of each task is analyzed, and the scheduling order of the computational tasks is determined based on the highest response ratio of the tasks; secondly, the optimal solution is determined by iterative search to offload the most suitable IoT tasks or computing tasks to the edge server. The experimental results show that this offloading strategy can effectively reduce the latency of computationally intensive edge network tasks and improve the resource utilization of the edge server. [ABSTRACT FROM AUTHOR]

The study systematically reviews the integration of Fog and Edge Computing within Learning Analytics to enhance data privacy and security in educational settings that use cloud computing. Employing the PRISMA methodology, we analyze current literature from Web of Science and Scopus databases to examine how these decentralized computing technologies can mitigate the risks associated with centralized cloud storage by processing data closer to its source. Our findings highlight the significant potential of Fog and Edge Computing to transform Learning Analytics by enabling real-time, context-aware data analysis that supports personalized learning while ensuring stringent data privacy. This approach challenges conventional data management practices, advocating for privacy by design in developing new strategies and frameworks. The research underscores the need for collaborative efforts in establishing standards and guidelines for secure and effective technology use in education, pointing towards the necessity of addressing technical, operational, and ethical challenges to maximize the benefits of fog and edge computing in Learning Analytics. • Fog and Edge computing improve data privacy for Learning Analytics and minimizes dependence on centralized cloud storage. • Nearer analytical technologies to users promotes a personalized and secure learning environment. • Future research should focus on creating secure architectures and policies to leverage Fog and Edge Computing in education. [ABSTRACT FROM AUTHOR]

Sudden medical safety accidents or large-scale outbreaks of epidemics will lead to a surge in traffic near hospitals and other medical infrastructure, and traditional edge base stations cannot cope with sudden traffic demands. In addition, the patient's data of health information is private, it is necessary to protect the patient's health information. Therefore, aiming at the practical problems that edge base stations cannot cope with a large number of communication requests and how to protect patients' private data under sudden medical safety accidents, this paper combines hierarchical federated learning with UAV-assisted mobile edge computing environment to build a UAV-assisted MEC system architecture for federated learning. However, federated learning is closely related to participating nodes. Without satisfactory rewards, users and UAVs will be unwilling to consume computing resources and communication resources to participate in federated learning. Therefore, this paper introduced the Stackelberg game and incentive mechanism into federated learning. The interaction between user devices, UAV, and the base station is modeled as a Stackelberg game to determine the maximum amount of data required for model training and the number of user devices and UAV. The incentive mechanism determines the task allocation and reward allocation of federated learning and determines the maximum benefits of three participants in the game process to improve the quality of user devices and the performance of the federated learning model. Experimental results show that the proposed algorithm can inspire user devices, which have high-quality data, to participate in federated learning training, improve training accuracy and maximize social welfare. [ABSTRACT FROM AUTHOR]

In the mobile edge computing system subject to wireless interference, the Edge Server Provider (ESP) aims to offer profitable computing resources to Device Managers (DMs), who make optimal strategies based on the provided prices. However, the existence of mixed variables typically constitutes an NP-hard problem, posing a significant challenge for optimization. In response to address this issue, we formulate a bi-level optimization problem, where the upper level is devoted to optimizing the pricing of computing resources. The lower level optimizes DM migration strategies and resource allocation at specified prices. Leveraging game theory, we achieve distributed and efficient computation offloading by formulating the distributed computing offloading strategy among lower-level DMs as a task offloading game. Our analysis identifies Nash equilibrium and finite improvement characteristics within the game. Based on these insights, we present a bi-level distributed computation offloading algorithm capable of reaching Nash equilibrium, thus optimizing the profit for both DMs and ESP. The experiments have demonstrated the algorithm's effectiveness in reducing system costs and maximizing the profits of ESP with DMs across diverse scenarios. [ABSTRACT FROM AUTHOR]

Mobile edge computing (MEC) is a promising technology to support computation-intensive tasks for mobile devices which are usually associated with limited resources. Many researches from both scientific and industrial field have put focuses on MEC. However, most of them assume that in a MEC environment, the offloaded tasks are independent or that there is only one server in the MEC center. Nevertheless, in reality, tasks with dependencies take the majority and in a MEC center, there are usually multiple servers. Under this circumstance, previous methods no longer take effects. In this work, we consider offloading with precedence constraints among tasks, and try to minimize makespan over a MEC center with multiple servers. This problem becomes more complex given that a task can not start unless its predecessors are completed. To solve the problem, we jointly involve front end task offloading order and back end scheduling to optimize makespan, and propose a corresponding algorithm called joint re-ordering and frequency scaling (JRFS). Extensive experiments have been conducted. The results show that compared with several other methods, JRFS can achieve better makespan. [ABSTRACT FROM AUTHOR]

Recent years have witnessed the proliferation of mobile computing and Internet-of-Things (IoT), where billions of mobile and IoT devices are connected to the Internet, generating zillions bytes of data at the network edge. Driving by this trend and the development of wireless communication, edge computing, an emerging computing paradigm, has received tremendous amount of interest. By pushing data storage, computing, and control closer to the network edge, edge computing has been widely recognized as a promising solution to meet the requirements of low latency, high scalability, prompt response, and energy efficiency. In the meanwhile, with the development of neural networks, Artificial Intelligence (AI) has been applied to a variety of disciplines and proved highly successful in a vast class of intelligent applications cross many domains. Edge intelligence, aiming to facilitate the deployment of neural networks on edge computing, has received significant attention. However, there are many challenges existing for a novel design of edge computing architecture to AI applications, and their co-optimization. For instance, conventional neural networks techniques usually entail powerful computing facilities (e.g., cloud computing platforms), while the entities at the edge may have only limited resources for computations and communications. This suggests that AI algorithms should be revisited for edge computing to AI models into the edge device for efficient processing. On the other hand, the adapted deployments of neural networks at the edge empower the efficient learning systems that can provide the "smartification" across different layers, e.g., from network communications to applications, and also involve collaborations across edge to cloud. Finally, designing algorithms for small-scale edge devices in a learning ambience is all the challenging as there are several conflicting issues to account for. These include, memory management, power management, and compute capability of a node, etc. [ABSTRACT FROM AUTHOR]

In the context of next-generation 5G and beyond communication networks, integrating Unmanned Aerial Vehicles (UAVs) with Mobile Edge Computing (MEC) is crucial. Hybrid Non-orthogonal Multiple Access (H-NOMA) has been recognized as a prominent technique for reducing energy consumption during data offloading. However, the literature assumes that all users in the cluster have latency requirements and interference levels such that implementing H-NOMA is optimal, overlooking other scenarios. Furthermore, the position of UAV-hosted MEC is not optimized. To address these constraints, we propose an adaptive offloading method where users can utilize either H-NOMA or OMA for data offloading in designated time slots based on their conditions. We substantiate this proposal through a comparative analysis of energy consumption between H-NOMA and OMA. Additionally, we introduce a novel Maximum Latency Difference Clustering and Power Allocation (MLDC & PA) algorithm for organizing smart terminals (STs) and allocating power. Furthermore, we propose a heuristic-based optimization approach for UAV positioning to minimize offloading energy and enhance network efficiency. Simulation results confirm that the proposed approach has superior energy consumption reduction capabilities compared to state-of-the-art techniques. • Employing Maximum Latency Difference-based Clustering to organize users effectively. • Implementing an Adaptive Offloading Method within each cluster to minimize energy consumption. • Strategically optimizing UAV altitude to enhance system performance. • Validating the superiority of the proposed method through comparative simulation with conventional Orthogonal Multiple Access (OMA) techniques. [ABSTRACT FROM AUTHOR]

With the swift advancement of 5G communication technology and the widespread prevalence of intelligent connected vehicles, a multitude of applications for the Internet of Vehicles (IoV) have been steadily burgeoning. Nevertheless, owing to vehicular mobility and the variety of tasks generated by IoV applications competing for restricted edge resources, the task of assigning users to edge nodes, in addition to appropriating edge node resources to hybrid tasks within edge environments, persists as a formidable challenge. In addressing this predicament, we establish a model delineating the execution duration of hybrid tasks engendered by IoV applications, and introduce a joint mechanism for user assignment and edge resource allocation. Within this mechanism, we architect an optimal strategy for edge resource allocation, predicated upon given user assignment outcomes. In order to ascertain the user assignment strategy, we employ an enhanced genetic algorithm in scenarios of a large user base, and utilize the branch-and-bound algorithm when dealing with a small user base. Finally, we design and implement an experimental setup for joint user assignment and edge resource allocation. The comprehensive simulation results provide robust verification of the effectiveness of our proposed mechanism. [ABSTRACT FROM AUTHOR]

The efficient and precise acquisition of hydrodynamic results in complex interaction scenarios for unmanned underwater vehicles (UUVs) presents a significant challenge. This study proposes a time-sequenced hydrodynamics prediction system of underwater vehicle based artificial intelligence edge computing. Firstly, a SE-CausalNet framework is proposed and a hydrodynamic prediction system for underwater vehicles is developed based on a neuromorphic computing chip. Secondly, the computational fluid dynamics (CFD) results of the Remus UUV are validated with experimental data. Two UUV motion cases are proposed to further elucidate hydrodynamics characteristics under the complex interference of submarines. Finally, the hydrodynamics of the UUV for two motion cases is analyzed, and four datasets are constructed to fully demonstrate the predictive capability of the system. Results show that the pressure field for the free offset path case differs remarkably from that of the fixed path case, and noticeable deviations are observed in both horizontal and vertical planes from the original fixed trajectory of the UUV. The prediction aligns well with CFD results, with a sample inference time within 0.3 s. The mean squared error of the four datasets is all below 2.6 × 10−4, indicating a rapid and highly accurate assessment of disturbance moments and trajectories. • The hydrodynamics characteristics of UUV during interactions with a submarine are investigated. • A hydrodynamics prediction system based on AI edge computing platform is developed. • It provides valuable insights to guide the design and motion control of UUVs. [ABSTRACT FROM AUTHOR]

A distributed collaborative optimal dispatching strategy for the integrated energy system (IES), based on edge computing and consistency algorithm, is proposed in this paper. To solve the optimal dispatching problem of the IES coupled with electricity, heat, and gas, a cloud-edge-device architecture is constructed. A distributed group consensus algorithm (DGCA) is designed with two consistency protocols. The first protocol defines three groups of consistency variables, represented by the incremental cost of electricity, heat, and gas, respectively. The output estimation deviation of each component is used as a feedback. The second protocol considers the output estimation deviation of the three components as the consistency variables. The coupling variables of the two consistency protocols tend to be consistent, which can obtain the optimal output of each unit and achieve the optimal operation of the system. Furthermore, the proposed strategy accounts for transmission loss during transmission. Finally, through the hardware in the loop simulation based on the edge intelligent terminal (EIT), the effectiveness and feasibility of the strategy are verified, considering different cases. • Edge computing technology to economic dispatch problem of IES. • A cloud-edge-device architecture is constructed. • Edge intelligent terminal (EIT) is designed for edge layer. • A distributed group consensus algorithm is designed with two consistency protocols. • Effectiveness and feasibility through hardware in the loop simulation. [ABSTRACT FROM AUTHOR]

As an important concept to depict the freshness of an information when it arrives at the destination, the Age-of-Information (AoI) and its relative scheduling works can be a very useful tool to aggregate the information from the end devices to the edge servers in Internet of Vehicles, which can strongly support those time-sensitive services in intelligent transportation system. However, most of the previous works about the AoI scheduling on edge computing are the centralized ones, rely on the reliable communications, and ignore the mobility of vehicles, which may not suit the dynamic environment in Internet of Vehicles. In this paper, we consider the AoI optimization problem in a more comprehensive vehicle model, and present a distributed AoI optimization algorithm based on a single hop end-to-edge wireless network. With an arbitrary power control technique, our algorithm reaches an asymptotically optimal performance on minimizing the expectation of AoI. Both of theoretical analysis and extensive simulations are conducted to verify the correctness and efficiency of our proposed algorithm. [ABSTRACT FROM AUTHOR]

• A mobile edge computing system integrating air, space, and ground is examined. • A two-layer optimization technique is proposed. • A PSO algorithm is used in the upper layer of the algorithm along with genetic algorithm operators to optimize the UAV's deployment position. • The greedy algorithm is applied to optimize the offloading of processing jobs in the lowest layer of the algorithm. • The proposed approach can also reduce the usual job response time of the system. Several Unmanned Aerial Vehicles (UAVs) offer ground equipment low-latency edge computing services. On the other hand, low-Earth orbit satellites offer all-encompassing cloud computing services. The officially stated joint optimization problem is a mixed nonlinear programming problem, leading to the development of a two-layer optimization technique. It is suggested that a particle swarm optimization (PSO) algorithm can be used in the upper layer of the algorithm along with genetic algorithm (GA) operators to optimize the UAV's deployment position. The greedy algorithm is applied to optimize the offloading of processing jobs in the lowest layer of the algorithm. The feasibility and effectiveness of the proposed strategy have been validated by numerous numerical simulation tests. The findings show that it is similar to other benchmark algorithms. The proposed approach can also reduce the usual job response time of the system. [Display omitted] [ABSTRACT FROM AUTHOR]

With the advancement of the 5G networks, edge computing (EC) assisted Internet of Things (IoT) based applications demand real-time computation and high-volume data-intensive services. Due to the heterogeneity and limited resources of the edge nodes (ENs), and dynamic resource demand of the IoT applications, it is challenging to place the IoT services into the available ENs by ensuring performance measurements on quality of services (QoS). In this paper, a novel quantum-inspired particle swarm optimization-based service placement (QPSO-SP) algorithm is proposed for EC environment. The QPSO-SP is intended to achieve desired service placement while optimizing throughput, energy consumption, delay, and computation load of the system. Quantum particle (QP) is designed to represent a complete solution for IoT service placement in an EC environment. Decoding of the QP is done by using a novel double-hashing technique. The fitness function uses throughput, delay, energy consumption, and load balancing parameters. Extensive simulation is performed and comparison is done with the standard existing algorithms. The parametric study, Taguchi method is conducted. The statistical analysis, ANOVA, followed by Friedman test is also done. The simulation results indicate that the proposed QPSO-SP outperforms existing works in terms of energy consumption, delay, throughput, and load balancing. • A Quantum-Inspired PSO is proposed to address IoT service placement. • Quantum particle is designed to represent complete IoT service placement solution. • The decoding of quantum particle is done using a novel double-hashing technique. • Fitness function considers throughput, energy consumption, delay and load balancing. • Simulation shows efficacy of the proposed work with statistical analysis. [ABSTRACT FROM AUTHOR]

Real-time and in situ water vapor (H 2 O) sensors are increasingly demanded to indicate thermochemical parameters of energy systems. As an accurate, fast and non-intrusive sensing technology, laser absorption spectroscopy (LAS) enables simultaneous measurement of gas concentration and temperature. Due to the complex spectral computation, real-time and rapid LAS signal processing are still challenging. Most state-of-the-art LAS sensors utilize high-level computing units to post-process laser measurements in real time, hindering their portable deployment in harsh industrial environments. To address the above challenge, we developed an online LAS gas sensor on edging computing platforms, named as LAS-on-Edge. The developed sensor achieves an online monitoring rate up to 62.5 Hz by deploying a spectral-informed neural network on an embedded device. Simultaneously, it reserves the raw kilo-Hz measurements offline for underlying dynamics understanding. The LAS-on-Edge sensor's compactness also enables its flexible and in situ integration. The sensor is firstly used to monitor a slow mist diffusion and accumulation process, exhibiting a high measurement accuracy with low spectra recovery residuals below 0.01. It is further applied to online monitor the H 2 O concentration and temperature variation during a lab-scale propane combustion process. It demonstrates a minimal temperature deviation of 1.86 % compared to the thermocouple reference in the flame. The sensor also successfully detects the fundamental flame pulsation at 5.65 Hz as well as high-frequency pulsations exceeding 200 Hz, thereby effectively reveals underlying flame dynamic behaviors. • A rapid LAS sensor on edge computing is developed for online water vapor measurement. • The sensor achieves both 62.5 Hz online monitoring and kilo-hertz offline analysis. • The on-edge sensor exhibits significant flexibility for its in situ integration. • The sensor effectively reveals the underlying combustion dynamics behaviors. [ABSTRACT FROM AUTHOR]

Structural health monitoring (SHM) provides real-time data on the condition and performance of infrastructure, enabling timely and cost-effective maintenance interventions, and hence enhanced safety and extended service life. The computer vision-based non-contact sensor has emerged as a promising alternative to conventional contact-type sensors for structural displacement measurement and SHM. Many of the currently reported vision-based structural displacement measurement systems typically temporarily set up a video camera from a distance to the structure. The collected images or videos are usually stored locally and post-processed offline to obtain structural displacement responses, which is cumbersome and limited to short-term SHM applications. The recent development of technologies empowered by the Internet of Things (IoT) and edge computing has enabled real-time video processing and analysis at the source, minimizing latency, reducing bandwidth requirements, and enabling prompt decision-making, thereby enhancing efficiency and responsiveness compared to traditional offline video recording and processing systems. In this paper, an edge computing vision-based displacement measurement system (EdgeCVDMS) is developed. Video recording, processing, and displacement response identification are entirely performed on an edge device integrated with the vision-based displacement tracking algorithm, thereby greatly reducing the amount of data transmitted to the cloud server. The feasibility and applicability of the developed sensing system are experimentally validated on a laboratory-scaled transmission tower structure. The proposed EdgeCVDMS is cost-effective, easily deployable, and of great potential to be applied for the condition assessment of a larger population of aging civil infrastructure. • This paper proposes a displacement measurement method based on edge computing. • Vision-based algorithm and edge computing device are used for measuring displacement. • AWS is used for data management and visualization. • Experimental validations are conducted to verify the accuracy of the proposed device. • Different influence factors, such as lighting conditions, angle and distance are considered. [ABSTRACT FROM AUTHOR]

• Open-source dataset: 3929 images and 12 k bounding-box annotations used in this study. • Deep learning model optimization using pruning and integrating re-parameterization module. • Assessing the effects of training multiple image resolution on the optimized YOLO-Spot model. • YOLO-Spot_M model achieves accuracy (+1.3 %) and mAP (+2.7 %) compared to YOLO-base model. • YOLO-Spot_M with half-precision gains 5X times inferencing speed on an edge device. Recent advancements in deep learning (DL)-based model optimization techniques have resulted in better weed identification accuracy. However, optimizing these models to identify weeds in images captured using small unmanned aerial system (UAS) has not been much explored. Moreover, leveraging the optimized model on resource constrained edge platform that could be easily integrated with UAS for real-time weed identification could be of significant advantage in developing precision aerial spraying weed management technology. Therefore, this study proposes YOLO-Spot model that is based on YOLOv7-tiny architecture, has been optimized and reconstructed to identify weeds amongst crop plants in aerial images and videos. The optimized model tends to use a smaller number of trainable parameters and reduced feature map sizes for weed identification. Most of the redundant convolutional layers along with feature map sizes have been reduced coupled with an integration of a novel module re-parameterized convolutional layer (RCL) within the neck component of the network. Furthermore, YOLO-Spot model has been trained on the three image resolutions, 320 × 320, 640 × 640 and 1280 × 1280, and has been named as YOLO-Spot_S, YOLO-Spot_M and YOLO-Spot_L, respectively. Out of all the variants, YOLO-Spot_M model has achieved significant prediction accuracy as compared to other variants and a denser layered model YOLOv7-Base. YOLO-Spot_M model utilizes over 75 % less parameters and 86 % reduced GFLOPs compared to YOLOv7-Base. As per the results, YOLO-Spot_M has outperformed YOLOv7-Base by achieving +1.3 % and +2.7 % overall accuracy and mAP(@0.5), respectively. The optimized architecture utilizes 4X less power (in W) when trained on a normal graphical processing unit (GPU). Moreover, converting YOLO-Spot_M to half-precision (FP16) for resource constrained device deployment (AGX Xavier), led to a +0.6 % accuracy and 5X faster weed recognition accuracy in aerial images and videos during inferencing. Based on the metrics obtained, YOLO-Spot_M model is recommended model that could be integrated with remote sensing technologies for site-specific weed management. [ABSTRACT FROM AUTHOR]

With the popularity of IoT (Internet of Things) applications, edge computing has received lots of attention. To meet data privacy protection requirements of edge nodes and cope with their unbalanced data distribution, federated learning (FL), a distributed learning framework, is widely used in intelligent edge computing applications. However, recent studies have shown that FL still suffers from privacy leakage problems, including membership inference, data reconstruction, etc. However, these studies mainly focus on the feature information of private data. In this paper, we concern the user-level label privacy in FL. We propose LDIA, a label distribution inference attack against FL in edge computing, exploring the possibility that an honest but curious cloud server can infer the proportions of samples per label in the edge user's private data. LDIA is inspired by the observation that parameter changes in the output layer of a model can reflect the label distribution of training data. We use a neural network to learn individual features of the output layer updates over different label distributions, and then perform inference from local models uploaded by users. Our comprehensive evaluation shows that LDIA is effective on various datasets in different settings, demonstrating the severe privacy leakage in FL-based edge computing. [ABSTRACT FROM AUTHOR]

Asynchronous Federated Learning (AFL) has been introduced to improve the efficiency of FL by reducing the latency of Machine Learning (ML) model aggregation, particularly in the Internet of Things (IoT) environment. Meanwhile, decentralized FL, such as leveraging blockchain and directed acyclic graph (DAG)-based ledgers, also has drawn much attention to the integration of FL owing to the security benefit against single-point-failure and Byzantine fault tolerant consensus. We observe that the inherent network asynchrony of DAG-based ledgers is beneficial for implementing asynchronous FL particularly in edge computing domains, even though there is no existing survey work that provides a fundamental overview of such integration. This paper surveys on the integration of asynchronous FL with DAG, which we call AFL-DAG, as a promising approach to realize the intersection of decentralized FL and asynchronous FL. Motivated by the lack of a concrete taxonomy of asynchronous FL and a global model, especially with decentralized FL, we introduce universally applicable terminologies and the extensive classification method of FL in terms of asynchrony. Based on the proposed taxonomy, we present a generic system model of AFL-DAG for edge computing applications. To provide a horizontal overview and cover fundamental concepts in AFL-DAG, we identify four critical design factors and the state-of-the-art solutions are discussed accordingly. Future directions to achieve a practical AFL-DAG are also highlighted. Finally, we explore the opportunities of AFL-DAG by investigating a popular edge computing application, the on-device crowdsourcing, and provide a high-level evaluation of AFL-DAG compared to blockchained-FL. [ABSTRACT FROM AUTHOR]

Low latency in the detection of objects in images is a fundamental aspect to provide an immediate response in different application scenarios, and to carry out the relevant actions. Deep learning-based models offer better results than traditional object detection techniques. As a result of this increased accuracy, the computational cost of these models is often very high. For this reason, several edge computing devices have emerged that perform inference quickly and efficiently due to the incorporation of hardware accelerators. In this work, different devices are evaluated using object detection algorithms based on deep learning. For this purpose, YOLOv3, YOLOv5 and YOLOX, with all their variants, have been used on an NVIDIA Jetson Nano, an NVIDIA Jetson AGX Xavier and a Google Coral Dev Board. For the evaluation of the models and devices, one of the most widespread datasets, MS COCO, is used. In addition, they have been evaluated using twenty different input sizes and three frameworks (Pytorch, TensorRT and Tensorflow Lite). From the data obtained, data can be extrapolated to other models such as YOLOv8. Additionally, the FPS/Power Consumption and FPS/Cost ratios are analyzed, as well as their feasibility in a real use scenario. As a result of this work, valuable recommendations are provided for projects where this technology is to be applied. • Quantitative comparison of the most popular edge computing devices. • Use of four object detection models (AI). • Actual power consumption analysis of the devices. • Actual application using the devices. [ABSTRACT FROM AUTHOR]

• RTAL is proposed to calculate the lodging area of large-scale farmland in real time. • RTAL method can be directly deployed on the UAV without offloading to the cloud. • RTAL method has been validated experimentally in Nvidia Jetson Xavier NX. • It takes 80 min for RTAL to generate the predicting results in 10 km2 farmland. Rice is a globally important crop that plays an important role in feeding more and more of the world's population as we cope with climate change and population growth. Rice lodging is one of the main yield-reducing factors in rice production, which also has a direct impact on rice quality and leads to harvest difficulties, so it is very important to obtain lodging data in time. Lodging assessment is a tedious task that generally requires a lot of time and labor due to the large area of land involved. This study proposes an edge computing method for rice lodging areas suitable for unmanned aerial vehicles (UAVs) without post-processing, called Real Time Rice Lodging Area Calculation Method (RTAL). The RTAL method is based on deep learning and photogrammetry, and can calculate large-scale farmland lodging areas in real time using only edge computing devices. Tested on the edge computing device Nvidia Jetson Xavier NX, when the ground resolution is 0.1 m, the fastest prediction speed can reach 14417.9 m2/s. When the operation time of a single sortie is 80 min, the predicted area can reach 10 km2. This method greatly improves the efficiency of prediction, and can provide real-time help in rice yield measurement or disaster damage assessment. [ABSTRACT FROM AUTHOR]