Search

The aluminum laminate pouch of pouch batteries is highly prone to deformation, which can cause various surface defects, thereby affecting their service life and potentially posing safety hazards. To address this problem, we propose an algorithm named YOLOv8-UCB for detecting surface defects in pouch batteries, which is based on the YOLOv8 model. First, while retaining the original network structure, we replaced the backbone of YOLOv8 with the UniRepLKNet feature extraction network to achieve a larger receptive field. Second, we constructed a distributed focal detection head CLLAHead, to better capture the features at different scales. Additionally, after each Concat layer in the head, we incorporated BiFormer attention mechanisms to enhance content-aware perception. Finally, we replaced the loss function with Slide Loss to address the issue of sample classification imbalance. The experimental results indicate that the improved algorithm achieved significant increases in precision, recall, mAP@0.5 and mAP@0.5:0.95, with improvements of 11.7%, 8.2%, 8.9% and 6.5% respectively, over the baseline model. In comparative studies against contemporaneous methodologies, the refined algorithm exhibited pronounced superiority in detection capabilities.

Deep learning is an effective method for hyperspectral image (HSI) classification, where CNN-based and Transformer-based methods have achieved excellent performance. However, there are some drawbacks to the existing CNN-based and Transformer-based HSI classification approaches: 1) CNN-based methods are deficient in showing the extraction of multiscale features and localized features owing to the fixed-size input patch. 2) the MHSA module ignores the interaction capability between multiple attention heads, which leads to insufficient feature fusion in various directions. 3) The weights of attention heads in various directions are disregarded in the MHSA and attention heads are simply concatenated horizontally. To address the above-mentioned limitations, a novel multihead interacted and adaptive integrated transformer (MHIAIFormer) with spatial-spectral attention, which integrates the respective advantages of convolutions and transformers is proposed in this study. A pyramidal spatial-spectral attention (PS2A) feature extraction module is adopted to efficiently capture the localized and multiscale feature information of HSI. The output of PS2A is then sent to the transformer encoder stage through a grouped multiscale cross-dimension embedding module, which includes additive self-attention using multihead interaction and MHSA with adaptive multihead merging to capture the long-range dependencies of the features. Extensive experiments on four datasets verify that our proposed approach achieves more satisfactory classification accuracy when compared with state-of-the-art models. The overall accuracy of the proposed model achieved 95.97%, 98.68%, 92.68%, and 99.49% on four datasets.

The detection and removal of Foreign Object Debris (FOD) on airport runway present significant challenges due to the small objects and the range of complex weather conditions that impact visibility and equipment efficiency. To address the problem, an FOD detection model using Swin Transformer (ST) enhanced YOLOv5 is proposed and a geolocation prediction model based on machine learning regression algorithms is introduced. Subsequently, an Unmanned Aerial Vehicle (UAV) is deployed to acquire a dataset with 74,737 images featuring 21 distinct types of objects. Furthermore, the model is trained using a comprehensive dataset of FOD on airport runway. By integrating a self-attention model with Convolutional Neural Network (CNN), the FOD detection model is formulated, yielding promising outcomes. The ablation tests demonstrate varying degrees of enhancement in the Mean Average Precision (mAP) value across each part of the model. Comparative analyses against established algorithms including YOLOv5, YOLOX, and YOLOv7 reveal superior overall performance and enhanced capability in detecting small objects. Notably, the proposed model exhibits lower relative accuracy decay when presented with diverse input data. Additionally, FOD geolocation prediction tests underscore the effectiveness of the machine learning-based regression algorithm, highlighting its substantial potential for practical implementation.

In recent years, research on high-speed or ultra-high-speed maglev systems has become increasingly popular, and different countries or teams have been committing to several different suspension technologies. Among them, due to the profound accumulation of research on the normal-conducting electromagnetic suspension technology, China is actively adopting this suspension technology to develop a new high-speed maglev train faster than $600~km/h$ , and this work is mainly on the basis of Transrapid (TR) trains and Shanghai Maglev Line (SHML). To evaluate the feasibility of this project, the speed limit of the TR maglev system was explored from the perspective of its suspension and guidance systems. The analysis process was carried out from two aspects of aerodynamic interference and track irregularity. As for aerodynamic interference, the aerodynamic and electromagnetic simulation models were established based on the actual maglev train. The anti-interference abilities of the suspension and guidance systems were analyzed from two aspects of steady-state and transient-state respectively. The results verified that the train could operate safely when its speed below $800~km/h$ and the cross-wind speed below $26~m/s$ . As for track irregularity, based on the actual track construction indicators of the SHML, the relationship between suspension gap fluctuation and the train speed was studied through a dynamic calculation model of the suspension control system. The results showed that the suspension system had a risk of track collision when the train speed was greater than $717~km/h$ . In summary, from the perspective of its suspension and guidance systems, the current TR maglev system on the SHML has an upper speed limit. The conclusion would be valuable for the development of ultra-high-speed maglev systems in the future.

State estimation has long been a fundamental problem in signal processing and control areas. The main challenge is to design filters with ability to reject or attenuate various disturbances. With the arrival of Big Data era, the disturbances of complicated systems are physically multisource, mathematically heterogenous, affecting the system dynamics via isomeric (additive, multiplicative, and recessive) channels, and deeply coupled with each other. In traditional filtering schemes, the multisource heterogenous disturbances are usually simplified as a lumped one so that the “single” disturbance can be either rejected or attenuated. Since the pioneering work in 2012 (Guo and Cao, 2012), a novel state estimation methodology called composite disturbance filtering (CDF) has been proposed, which deals with the multisource, heterogenous, and isomeric disturbances based on their specific characteristics. With CDF, enhanced antidisturbance capability can be achieved via refined quantification, effective separation, and simultaneous rejection and attenuation of the disturbances. In this article, an overview of the CDF scheme is provided, which includes the basic principle, general design procedure, application scenarios (e.g., alignment, localization, and navigation), and future research directions. In summary, it is expected that CDF offers an effective tool for state estimation, especially in the presence of multisource heterogeneous disturbances.

False data injection attacks (FDIAs) against the load frequency control (LFC) system can lead to unstable operation of power systems. In this paper, the problems of detecting and estimating the FDIAs for the LFC system in the presence of external disturbances are investigated. First, the LFC system model with FDIAs against frequency and tie-line power measurements is established. Then, a design procedure for the unknown input observer (UIO) is presented and the residual signal is generated to detect the FDIAs. The UIO is designed to decouple the effect of the unknown external disturbance on the residual signal. After that, an attack estimation method based on a robust adaptive observer (RAO) is proposed to estimate the state and the FDIAs simultaneously. In order to improve the performance of attack estimation, the $H_{\infty}$ technique is employed to minimize the effect of external disturbance on estimation errors, and the uniform boundedness of the state and attack estimation errors is proven using Lyapunov stability theory. Finally, a two-area interconnected power system is simulated to demonstrate the effectiveness of the proposed attack detection and estimation algorithms.

Graph convolutional neural networks have received a lot of attention in various tasks dealing with graph data by aggregating information from neighboring nodes and passing node information. Many recent studies have looked at the impact of topological features on node classification tasks by altering aggregation based on degree values or incorporating topological data analysis into graph convolutional neural networks; however, graph data itself has many topological characteristics in complex networks. In most circumstances, the graph’s topological characteristics reveal the nodes’ similarity and facilitate the node classification task. This paper proposes a topological structure feature extraction method based on the concept of complex topological characteristics, which can obtain deeper topological features in the graph structure and use node features to obtain feature information that is more important for classification from both feature spaces. Evidence from experimental studies has established that the topological structure obtained by the method in this paper can be used as input to the GCN,and good results can be achieved on the classification task even without any external information on the nodes. In the graph dataset of connected topologies, the method exhibits a very large increase in accuracy and macro F1-score when compared to the state-of-the-art baseline model after mixing the node features.

Using deep learning to maximize the benefits of a series of risks in the securities market was a very interesting and widely concerned problem. From the perspective of multi-source driving, we proposed a feature combination method based on prior knowledge and designed a deep hierarchical strategy model to solve this challenge. This model includes a pre-judging module, which implements trend judgment of the time series, and an operation module that performs the trading actions. For the pre-judging module, we designed a regression constraint Wasserstein generative adversarial network (WGAN) to complete the mission instead of the common discriminant methods in the past. As for the operation module, in order to make the machine learning more compatible with human learning characteristics and having the ability to control risks, we designed a network structure based on deep deterministic policy gradient (DDPG), which is one of the reinforcement learning algorithms to decide continuous trading position. We used the Dow Jones Industrial Index and the Shanghai Stock Exchange Index to conduct training and testing. The results show that the proposed model has a good performance in terms of return on investment and robustness.

In 5G mobile networks, the convergence of cloud computing and communication leads to mobile edge computing, benefiting vehicular networks. However, the advent of a wide variety of new services and devices has changed the vehicular network landscape, challenging vehicle-to-network services’ migrations. In this paper, we focus on optimizing long-term average latency of multiple services with a different quality of services (QoS). We first introduce an offline algorithm, which can be used to find the optimal migration strategy of services. Then, we analyze the negative effect of trajectory prediction and suggest an optimizing method to reduce this effect by partial updating. Finally, based on this method, we propose a partial dynamic optimization algorithm to approximate the optimal solution, by integrating the priority queue, which utilizes QoS information. We simulate the average service latency, confirming that the proposed partial dynamic optimization algorithm keeps a stable service latency and performs better than other existing algorithms, considering the negative effect of trajectory prediction. We also verify that the proposed algorithm can meet the low latency requirement of the vehicles and the different demands of different services. Besides, the partial dynamic optimization algorithm has a lower time complexity.

As statistics show that most threats to information security in Internet of Things (IOT) are caused by data leakage, lots of methods have been developed to address the problem of data leakage prevention (DLP). However, most of these methods do not work well when the confidentiality of data changes frequently. We propose an Adaptive Feature Graph Update model (AFGU) to solve the problem by mapping the features of confidential data to the feature graph. First, the feature graph are built to record the features of confidential data which involve the sensitive terms and their context. Then, the improved evaluation method for the importance of each term is employed to update the feature graph according to the importance degree of each term. Finally, the confidentiality of data are determined by matching the features of the data with the feature graph. Experiments results show that the proposed method can detect confidential data effectively and efficiently.

In the Internet of Things, data sets gathered from sensor nodes are missing a significant fraction of data, owing to noise, collision, unreliable links, and unexpected damage. This phenomenon is more serious in some scenarios, which limits the applications of sensor data. It is therefore necessary to develop methods to reconstruct these lost data with high accuracy. In this paper, a new multi-attribute missing data reconstruction method based on adaptive weighted nuclear norm minimization is developed, and a K-means clustering analysis is embedded into this forecasting process to improve the prediction accuracy. First, to ensure that sensors in one group have similar patterns of measurement, we use a traditional machine learning algorithm, called K-means clustering algorithm to separate sensors into different groups. Second, considering the correlations among multiple attributes of sensor data and its joint low-rank characteristics, we propose an algorithm based on the matrix rank-minimization method of automatic weighted nuclear norm minimization, which adaptively assigns different weights to each singular value simultaneously. Moreover, we use the alternating direction method of multipliers to obtain its optimal solution. Finally, we evaluate the proposed method by using a real sensor data set from the Intel Berkeley Research Laboratory with two missing patterns, namely, random missing pattern and consecutive missing pattern. The simulation results prove that our algorithm performs well even with a small number of samples, and it can propagate through a structure to fill in large missing regions.

In this paper, we propose a novel quadrature phase-shift keying (QPSK) vector millimeter-wave (mm-wave) signal generation based on odd times of frequency without precoding. The radio frequency (RF) vector signal generation module is universal for the generation of millimeter-wave signal with any odd frequency multiplication. Instead of phase precoding, the phases of electrical mm-wave signal are multiplied by the frequency multiplication number at the receiver after carrier phase estimation. The frequency tripling, quintupling, and septupling schemes are achieved by simulation, and bit-error-ratio curves demonstrate that the proposed QPSK vector mm-wave signal generation scheme works well.

In vehicular ad hoc networks (VANETs), the vehicles will periodically exchange information via broadcasting security-related message, such as the states of vehicles and roads, in which the performance may become worse when the vehicle density increases. To improve the poor performance of network when vehicular density is high, an adaptive back-off scheme based on improved Markov model is proposed in this paper. First, an improved Markov model with a contention window ( $CW$ ) inheritance mechanism is applied to reduce the network load. The vehicle with the improved model will take beacon expiration into account and actively discard expired beacons to relieve the pressure on the channel. Then, an initial $CW$ selection mechanism is designed to improve the efficiency of accessing the channel through providing the optimal initial $CW$ for the vehicles with different densities. Finally, we define a network busy factor to describe the condition of broadcast channel and propose an adaptive back-off algorithm which can dynamically adjust the $CW$ according to the network busy factor. Simulation results show that the new scheme can significantly improve the performance of broadcast.

To make up the disadvantages of OFDM-QAM applied in 4G systems, some new non-orthogonal asynchronous PHY multicarrier techniques have been proposed for multiple 5G application scenarios, such as universal-filtered multicarrier (UFMC) and filterbank-based multicarrier (FBMC). Accurate classification of multiple multicarrier waveforms is indispensable in the future wireless communication systems. Here, we propose a new multicarrier waveforms classification system, which utilizes deep convolutional neural networks to classify OFDM-QAM, UFMC, and FBMC-OQAM. Since the classification accuracy is greatly reduced in AWGN environment, principal component analysis (PCA)-based processing method is proposed in this paper to suppress the AWGN and reduce the dimensions of inputs of convolutional neural networks. Amplitude data is used in this paper as input for CNN and shows promise when compared to which uses raw I/Q data as input. The dense channel environment effects including timing errors, carrier frequency offsets, and multi-fading channels are also considered in this paper and our proposed system can obtain the high classification accuracy in noisy channels.

This paper reports a monolithic integration of GaN high-electron-mobility transistor (HEMT) and green light-emitting diode (LED), where the circular HEMT is surrounded by a ring-shaped LED and two devices are seamlessly interconnected by the LED's n-GaN layer and the HEMT's two-dimensional electron gas (2DEG) channel. By adopting such a novel circular layout design, the green HEMT-LED shows a controllable and uniform green light emission at 507 nm by simply tuning its gate voltage. This enables a uniform, controllable green LED light source, serving as an essential element in the red-green-blue (RGB) LED solution for a wide range of applications, such as tunable-spectrum white LED illumination, multichannel visible light communication with wavelength division multiplexing, RGB-based full-color LED displays, and optogenetics.