The inherent challenges of small objects in remote sensing imagery encompass the degradation of fine-grained spatial details throughout the downsampling stages, semantic inconsistency during multi-level feature fusion, along with unreliable localization caused by noisy samples. To address these issues, this paper proposes an efficient small-object detector termed EMW-YOLO. An Efficient Down-sampling (EDS) module is introduced to preserve fine-grained spatial information and enhance feature representation during feature extraction through spatial rearrangement and cross-dimensional attention. A Multi-Scale Fusion and Enhancement (MSFE) architecture is further developed to improve semantic consistency across feature levels by combining local enhancement with global feature alignment. In addition, the WIoU v3 loss function is also incorporated to suppress outlier samples and enhance the robustness of regression for small objects. Experimental results on the VisDrone2021 dataset demonstrate that EMW-YOLO achieves absolute improvements of 6.2%, 6.9%, and 7.9% over YOLOv8n in precision, recall, and mAP0.5, respectively. Further evaluation on the NWPU VHR-10 and RSOD datasets confirms that the proposed method exhibits strong generalization capabilities.

To alleviate communication pressure and terminal resource constraints in mobile edge computing (MEC) networks, this paper proposes a resource allocation optimization method for MEC systems that integrates data compression technology and non-orthogonal multiple access technology. This method considers practical constraints such as terminal device battery capacity and computational resource limitations. By jointly optimizing computational resource allocation, task offloading strategies, and data compression ratios, it constructs an optimization model aimed at minimizing the total task processing latency. Addressing the challenges stemming from the non-convex nature of the problem and the dynamic variations in channel conditions and task requirements, this paper proposes a softmax deep double deterministic policy gradient algorithm, where softmax operator function mitigates both overestimation and underestimation biases inherent in traditional reinforcement learning frameworks, enhancing convergence performance. Utilizing a deep reinforcement learning framework, the algorithm achieves joint decision-making optimization for computational resources, task offloading, and compression ratios, thereby minimizing the total task processing latency while satisfying transmit power and computational resource constraints. Simulation results demonstrate that the proposed scheme exhibits significant advantages over benchmark algorithms in terms of convergence speed and task processing latency.

In complex electromagnetic environments, the scale effect of Unmanned Aerial Vehicle (UAV) swarm presents significant potential for enhancing cooperative effectiveness. However, the accuracy of Time Difference of Arrival (TDOA)-based localization for non-cooperative emitters using UAV swarm is significantly affected by the coupling of multi-source errors, which mainly include UAV position error (UPE), clock synchronization error (CSE), and TDOA measurement error (TME). To address the challenges of evaluating cooperative effectiveness under multi-source errors coupling and balancing localization accuracy with computational efficiency, a cooperative utility of information (CUoI) optimization approach is proposed.First, a TDOA observation uncertainty model is constructed by integrating multi-source errors. Then, the information gain of target position estimation is derived to build the CUoI evaluation model. Next, the characteristic of Dueling Deep Q-Network (Dueling DQN) that decouples state value from action advantage is leveraged, enabling precise evaluation of the potential benefits of different hyperparameter adjustment strategies. This characteristic facilitates adaptive tuning of key hyperparameters in Particle Swarm Optimization (PSO). Finally, a dynamic PSO framework based on Dueling DQN is proposed to effectively balance localization accuracy and computational efficiency. Numerical experiments demonstrate that the proposed algorithm achieves reductions in average localization RMSE of 19.1%, 6.0%, and 1.4%, respectively, compared to Semidefinite Relaxation-TDOA (SDR-TDOA), Grey Wolf Optimizer (GWO), and Multi-swarm Discrete Quantum-inspired Particle Swarm Optimization with Adaptive Simulated Annealing (MDQPSO-ASA).

In passive sensing, it is crucial to develop direction-of-arrival (DOA) estimation approaches that do not depend on knowing the number of sources in advance. Wireless signals are often disrupted by impulsive noise during propagation, which undermines the performance of traditional DOA estimation approaches that assume stationary Gaussian noise. To address these challenges, we propose a robust sparse DOA estimation algorithm based on gridless variational Bayesian theory. This approach models noise as a mixture of Gaussian and generalized t distributions within a hierarchical Bayesian framework. It incorporates a noise weight factor to adapt to various noise environments. The method updates the posterior probabilities of hidden variables and model parameters using variational Bayesian inference, providing an uncertainty measure for DOAs. Experimental results show robust performance and high accuracy under low signal-to-noise ratio (SNR) conditions and limited samples.

In disaster areas, Flying Ad-Hoc Network (FANET) is a critical method for emergency communication. However, the limited payload capacity of Unmanned Aerial Vehicle (UAV) leads to communication link interruption and routing congestion, thus it is a great challenging to the deterministic routing of FANET. This paper proposes a Q-learning based multi-objective optimization routing strategy (QMR) in the UAVs deterministic network. The strategy considers delay, bandwidth and energy as QoS measurement to select a main path. Moreover, the UAV node can sense the energy of next node in the main path. When the energy of next node is low, the current node calculates a disjoint backup path. The simulation results show that compared with the existing routing methods, our strategy can provide higher packet reception rate, lower average end-to-end delay and lower energy consumption.

With the widespread adoption of mobile robots, path planning and intelligent navigation have become hot research topics. In order to enhance the globality of traditional robot path planning algorithms, a deep Q-learning network (DQN) algorithm is proposed. By utilizing the reinforcement learning method, mobile robots can effectively navigate to the target location while avoiding the problems in conventional path planning algorithms, such as redundant pathways, decreased continuity and reliance on local information. The proposed method results in a significant reduction in inflection points of the inspection path, mitigating the occurrence of local optima and providing a highly optimized global solution for path planning under known map conditions. In conclusion, the proposed algorithm has been simulated and compared with conventional path planning techniques utilizing a two-dimensional raster map. Empirical findings attest to the dependability of the proposed global optimization technique, which has culminated in the generation of an optimized planned path.

In recent years Non-Intrusive Load Monitoring (NILM) technology has developed rapidly, with applications covering power management, smart homes, fault detection, equipment condition monitoring and other areas. In commercial building power inspection scenarios, traditional inspection methods often suffer from low efficiency, high costs, missed inspections and false inspections. At present, the use of NILM technology can identify the working status of each power equipment, significantly reducing the probability of missed and false detection. However, there is still the problem of not being able to balance classification efficiency and classification accuracy. In this study, a high frequency fast non-intrusive load monitoring system is designed. The use of high-frequency data sets significantly reduces the data acquisition time. To further reduce the time cost, an ensemble learning algorithm based on K-Nearest Neighbor (KNN) and Quadratic Discriminant Analysis (QDA) is proposed. The algorithm not only achieves an F1-score of 90

With the rapid growth of internet traffic, network congestion becomes more and more severe, which causes massive packets loss. The reliability and delay of data transmission needs to be guaranteed in real-time applications such as financial transactions and cloud games. Traditional transmission strategies use packet retransmission to ensure data reliability, but the retransmission causes extra delay. The extra delay reduces the quality of service (QoS) for deterministic services. For the above problem, this paper proposes a deterministic network (DetNet)-oriented multipath transmission strategy in the software-defined network (SDN) architecture. The architecture introduces the packet replication and elimination function (PREF) of DetNet to achieve reliable transmission. The strategy establishes a path optimization model with transmission delay and packet loss rate, and then solve the model by the Q-learning algorithm. The delay and packet loss rate of the link construct the reward function. The Q-value table enables to obtain the best combination of paths, and it is gained by the reward function. Simulations show that our strategy have lower packet loss rate and delay jitter than the traditional single-path transmission strategy and the multi-path transmission strategy.

The time difference of arrival (TDOA) multi-target localization based on unmanned aerial vehicle (UAV) swarm faces challenges such as limited localization accuracy and complex coordination due to uncertainty biases such as node position errors and perceptual performance differences (NPEAPD) for nodes. The localization error is directly related to the spatial geometric configuration formed by the nodes and target. Therefore, node selection for localization becomes a key factor in improving the accuracy and efficiency of localization. In this paper, aiming at the difference in node perceptual performance, we use prior observation information of the targets to derive the lower bound of TDOA perceptual performance and analyze the key influencing factors of localization performance. To address the problem of limited performance of node selection in traditional TDOA localization under the aforementioned uncertain biases, we introduce a node selection strategy based on localization basics and contribution. We measure the localization basics and contribution using the received signal-to-noise ratio (SNR) of the nodes and the expected information about target parameters provided by the localization group, and construct an optimization model to solve it with DMSSA algorithm. Simulation show that the proposed node selection strategy has significant performance advantages compared to three benchmark strategies and improved localization accuracy.

Recently, Consistency-based Metric Generative Adversarial Networks (CMGANs) have been proposed for the field of noise suppression, as they are capable of capturing local and global dependencies. However, CMGAN algorithms are designed for high signal-to-noise ratio environments. When applied to commonly encountered low signal-to-noise ratio environments in outdoor settings, factors such as masking effects lead to decreased feature extraction performance, resulting in a decline in denoising effectiveness. This paper introduces an improved multi-discriminator CMGAN noise suppression algorithm. The algorithm employs multiple feature inputs to enhance the information content of audio inputs and utilizes multiple discriminators to differentiate based on different frequency components. This enables it to learn features specific to different frequencies, thus achieving favorable noise suppression results even in low signal-to-noise ratio environments. For instance, simulation results on the TIMIT dataset and the Google environmental sound dataset demonstrate that the proposed algorithm, in comparison to the original CMGAN, achieves improvements in PESQ and STOI scores in low signal-to-noise ratio conditions, with an average improvement of 4.6% and 2.4%, respectively.

Low Earth orbit(LEO)satellite systems provide terrestrial users with services that are not lim-ited by geographical location.However,the conflict between existing allocation schemes and the business variability between beams is becoming increasingly prominent.Beam hopping technology allows for a more flexible and versatile approach to satellite re-source allocation.This paper proposes a beam hop-ping pattern optimization scheme that jointly consid-ers the interference threshold distance and beam ser-vice priority,reducing the inter-beam co-channel in-terference(CCI).In the cluster area,a non-orthogonal multiple access(NOMA)-based collaborative beam hopping(NCBH)scheme is proposed to minimize the cell-edge user(CEU)interference.Since there is a difference in channel gain between the CEU and cell-center user(CCU),this scheme forms a NOMA cluster to perform power domain multiplexing and formulates a NOMA cluster pairing strategy according to the user location to reduce the CCI of the CEU.After NOMA cluster pairing,the optimal carrier frequency of the NOMA cluster is selected by a reinforcement learning algorithm.The simulation results verify the excellent performance of the proposed NCBH scheme regard-ing the user's received power,transmission rate,and outage probability.

In low earth orbit (LEO) constellations, channel utilization is low owing to variations in the traffic distributions between beams. LEO satellites improve channel utilization via the full-band multiplexing technology. However, such satellites produce significant inter-beam interference. Additionally, this inter-beam interference becomes more aggravated owing to the requirements of multiple satellite coverage. To resolve this issue, we propose a LEO channel allocation scheme based on an improved artificial bee colony (IABC) algorithm. This scheme considers the difference in inter-beam communication traffic and the limitation of co-channel interference. A binary integer-programming model and IABC algorithm allocate channels to maximize the system throughput. The channel allocation matrix corresponds to a feasible solution in the IABC algorithm. The simulation results verify the excellent performance of the proposed scheme in terms of the throughput, blocking rate, and propagation delay compared with the performance of traditional schemes.

随着时代发展,群智感知技术在各行业中的应用范围不断增大,而具有低成本、智能化等特点的无人机也逐步走向集群化,蜂群定位成为基础设施不完备场景的新型解决方案.为了减少蜂群协同复杂度,可通过选站策略等对协同数量进行优化.但对于运动目标,在提高定位性能的同时,频繁选站会导致运算复杂度倍增,即存在定位性能与定位复杂度之间的矛盾.针对上述问题,设计出了一种基于扩展卡尔曼滤波的马尔科夫修正交互式多模型跟踪算法,在增加目标运动跟踪算法的基础上,减少选站次数,并通过引入交互式多模型算法对多种运动模型进行适配,以弥补单模型算法的缺陷.同时,在蜂群定位场景下对交互式多模型的转移概率进行自适应性更新,提高模型匹配度,实现对目标真实运动轨迹的跟踪预测.由对比实验结果可知,该算法可大幅缩短模型切换时间,从 10～20 s缩短至 5 s,降低了定位复杂度.

In the ultra-wide band (UWB) indoor scenario, which is affected by the multipath effect, the rational selection of positioning base stations to provide high-precision indoor location services is a hot research topic nowadays. Typically, base station selection is based on link classification with less consideration of base station configuration. In this paper, the Signal Interference Noise Ratio (SINR) of the received signal and the Position Dilution of Precision (PDOP) factor are taken into account to optimize the base stations participating in the positioning computation. The ultimate goal is to increase the positioning accuracy and improve the spatial distribution of the positioning error. Experiments prove that in a $33\mathrm{m}^{\ast}5.1\mathrm{m}^{\ast}3.4\mathrm{m}$ indoor showroom, the algorithm in this paper is compared with the base station selection method based on SINR and the method in which all base stations participate in the localization, etc., and the localization accuracy is improved and the localization results have a certain degree of stability.

UAV swarms are widely used in radar signal interception due to their advantages of wide sensing range and rapid information sharing. Aiming at the problem that the signal samples intercepted by UAV cluster are difficult to be fused and analyzed directly, and the recognition accuracy of multi-function radar（MFR） working mode is low under the condition of few training samples and unbalanced working mode samples, an MFR working mode recognition method based on smooth graph signal generated by self-organizing map（SOM） clustering is proposed. Firstly, the intercepted signal samples are clustered by using distributed SOM algorithm to extract the similarity between samples; Then, according to the clustering results, the signal sample set is characterized by smooth graph signal, and the correlation of signal samples under the same working mode is established; Finally, the graph attention network is used to fuse and classify the graph node data of the above graph signals to complete the MFR working pattern recognition. The experimental results show that, when the imbalance of working mode samples is about 10∶1 and the number of training samples in each class is 25, the recognition accuracy and F1 measure of this method are improved by 22.8% and 22.34% respectively compared with the existing methods, and can be applied to the case of noise interference.

Environmental sound intelligence detection is one of the important means of smart city construction. The performance of the detection is particularly affected by environmental noise. To increase the performance of environmental sound (ES) recognition, different target features and masks are used in denoising, which has large differences in numerical range. However, as the number of targets with large numerical differences increases, the difficulty of network learning data rules will increase, and so will the target estimation error. The multi-objective learning noise suppression algorithm put forth in this paper aims to reduce the influence, based on a convolutional and dual temporal convolutional neural network (CNN-2TCN). The algorithm combines CNN and two parallel TCN modules to learn two types of targets with large differences in numerical range, which can optimize the training process of neural networks. In addition, the secondary targets and the IRM-based adaptive post-processing method are adopted to improve the prediction accuracy of primary targets and the quality of the reconstructed ES. The experimental results show that the proposed algorithm achieves better performance than other algorithms. Moreover, the proposed algorithm can improve the performance of the ES detection system. Especially, the ES detection system, which has used the proposed method, has made an increase of 5.95%, 31.35%, and 36.75% in the recognition accuracy under the three signal-to-noise ratios (SNR) of 5 dB, 0 dB, and -5 dB, respectively.

Low earth orbit (LEO) satellite communication networks require huge load capacity and information processing speed to carry global communication traffic. Inter-satellite links and the on-board processing are the key technologies to achieve this goal, but the new network architecture leads to great challenges on satellite routing. This paper designs a hybrid inter-satellite link with the same-orbit laser and the different-orbit microwave to increase the link capacity and adopts a CPU centralized scheduling to improve the utilization of computing resources. Then, this paper establishes minimum delay function by considering the inter-satellite transmission delay and the on-board processing delay. The transmission delay model bases on the orbital period, and the processing delay adopts the multi-services model, the limited-capacity single-service model, and the unlimited-capacity single-service model in the queuing theory to model the on-board CPU centralized scheduling, photoelectric converters, and electro-optical converters, respectively. Based on this model, this paper proposes an inter-satellite routing strategy with modified Q-routing algorithm. The modified algorithm uses Dijkstra algorithm to accelerate the convergence of Q-routing algorithm and retains the strong real-time performance of Q-routing algorithm. Simulations show that the delay of the modified algorithm is 83.3 $$\%$$ % lower than that of the Dijkstra algorithm, and the larger the network and the traffic, the more obvious the advantage.

Mobile edge computing (MEC) is considered to be a promising technique to enhance the computation capability and reduce the energy consumption of smart mobile devices (SMDs) in the sixth-generation (6G) networks. With the huge increase of SMDs, many applications of SMDs can be interrupted due to the limited energy supply. Combining MEC and energy harvesting (EH) can help solve this issue, where computation-intensive tasks can be offloaded to edge servers and the SMDs can also be charged during the offloading. In this work, we aim to minimize the total energy consumption subject to the service latency requirement by jointly optimizing the task offloading ratio and resource allocation (including time switching (TS) factor, uplink transmission power of SMDs, downlink transmission power of eNodeB, computation resources of SMDs and MEC server). Compared with the previous studies, the task uplink transmission time, MEC computation time and the computation results downloading time are all considered in this problem. Since the problem is non-convex, we first reformulate it, and then decompose it into two subproblems, i.e., joint uplink and downlink transmission time optimization subproblem (JUDTT-OP) and joint task offloading ratio and TS factor optimization subproblem (JTORTSF-OP). By solving the two subproblems, a joint task offloading and resource allocation with EH (JTORAEH) algorithm is proposed to solve the considered problem. Simulation results show that compared with other benchmark methods, the proposed JTORAEH algorithm can achieve a better performance in terms of the total energy consumption.

本文分析了两站和三站时差分选的性能和运行效率,提出了一种基于云模型的辐射源信号多站分选方法.该方法首先利用两站时差分选方法完成脉冲配对与粗分选;然后运用云模型计算各配对成功脉冲集合在时差、脉宽和载频参数维度上的隶属度,合并来自同一辐射源的脉冲集合,实现细分选;最后利用另一个云模型计算配对失败脉冲与细分选后的集合在脉宽、载频和带宽参数维度上的隶属度,归类配对失败脉冲.仿真结果表明,与现有多站时差方法相比,本文通过引入云模型合并时差粗分选结果中来自同一辐射源的脉冲集合,降低了虚警率,同时实现了配对失败脉冲的准确归类,提高了方法的鲁棒性.

The demand for an indoor localization system is increasing, and related research is also becoming more universal. Previous works on indoor localization systems mainly focus on the acoustic signals in Line of Sight (LOS) scenario to obtain accurate localization information, but their effectiveness in Nonline of Sight (NLOS) scenario remains comparatively untouched. These works are usually less efficient as the acoustic signals often bring diffraction, refraction, scattering, energy decays, and so on in NLOS environments. So the system needs adjusting accordingly in a complex NLOS scenario based on NLOS identification results. Therefore, the identification of NLOS acoustic signal turns out to be significant in the indoor localization system. If the system only uses original support vector machine (SVM) to complete NLOS identification, the result turns out to be poor by our test. To address this challenge, we propose a novel indoor localization system, named ZKLocPro, which utilizes an advanced swarm intelligence method to optimize the traditional SVM classification model to deal with NLOS acoustic signal identification. Its results can help the system adjust the localization process if necessary in a complex NLOS scenario. Obviously, it is also significant to build our own NLOS data set, which is suitable for an indoor localization system’s situation. Specifically, four methods are added: (1) new LOS and NLOS acoustic localization signal sample production, rearrangement, and reselecting process; (2) advanced parameter optimization process; (3) elitist strategy; and (4) inertia weight nonlinear decrement. The experimental result shows that our system is efficient and performs better than state-of-the-art congeneric works even in a complex NLOS scenario.