最新刊期
- 摘要：To address resource competition and performance assurance in hybrid wireless and power-line communication networks with coexisting single- and dual-mode nodes and heterogeneous traffic, a MAC algorithm based on adaptive superframe parameters and multi-agent resource coordination is proposed. A two-stage superframe protects single-mode access while enabling intelligent competition among dual-mode nodes. As a longer protection stage improves single-mode performance but reduces dual-mode resources, a dynamic algorithm combining offline baseline selection and online calibration determines its length and adjusts the MARL subslots and action space accordingly. Dual-mode access in the second stage is formulated as a joint link-selection and slot-allocation problem, and a multi-agent reinforcement learning model enables coordinated access through link-state awareness and intelligent parameter optimization. Simulations show that the algorithm satisfies the throughput and head-of-line (HOL) delay constraints of single-mode nodes under varying loads and network sizes. Under heavy load with 18 dual-mode nodes, it improves aggregate throughput by 17.8% over DCMA and reduces the MARL-stage collision probability by 35%.关键词：dual-mode hybrid access;power line communication;MAC access;multi-agent reinforcement learning;pa- rameter adaptation3|0|0更新时间：2026-10-10
- 摘要：Aiming at the problem that the ultra-wideband (UWB) positioning system is affected by non-line-of-sight (NLOS) propagation in complex indoor scenes, resulting in insufficient ranging accuracy and degraded positioning performance, an adaptive fusion network guided by physical features was proposed. Channel impu-lse response (CIR) was adopted as the processing object, and a combination of time-domain and frequency-domain analyses was employed for signal processing : First, the fast Fourier transform was performed on the CIR signal to extract the frequency domain features, and the physical features such as the first-path energy ratio, peak comparison, and multipath expansion were extracted. Then, the dynamic fusion of time-frequency domain features is completed by using physical feature guided bidirectional gated recurrent unit ( BiGRU ) and channel attention mechanism, thereby improving the model’s capability to distinguish NLOS signals in complex environments. The experimental results show that the comprehensive performance of PGAF-Net on public data sets is better than that of comparison algorithms, and it has better accuracy and stability in NLOS signal recognition.关键词：ultra-wideband (UWB);non-line of sight (NLOS);channel impulse response;dynamic integration;attention mechanism8|2|0更新时间：2026-09-29
- 摘要：As embodied intelligence evolves from individual autonomy to collective collaboration, the embodied swarm-agent system demonstrates collective intelligence that surpasses the sum of individual capabilities through multi-agent information interaction and cooperation in complex open environments. However, existing mobile communication networks and Internet of things (IoT), which focus on enhancing the sum rate and the number of connected IoT devices, respectively, struggle to support the loop "perception-interaction-decision-action" based embodied swarm-agent system in collaboratively executing tasks within open physical environments. Therefore, investigating the task-driven embodied swarm-agent network, where embodied agents serve as nodes, represents a core direction for realizing embodied swarm-agent systems. Firstly, the core characteristics and key technologies of embodied swarm-agent systems were presented, and their key requirements for communication networks were extracted, including physical-world task orientation, high-value semantic interaction, agile topology reconfiguration, and sensing-communication-computation-control coordination. Secondly, the development of mobile communication networks and IoT was reviewed, and their limitations in supporting embodied swarm-agent systems were analyzed. On this basis, the embodied swarm-agent network was proposed to support embodied swarm-agent systems. Its core features were defined, including semantic-driven lightweight communication, task-centric network scheduling, self-organized dynamic topology management, and resource-adaptive coordination enhancement of sensing, communication, computation, and control. For the embodied swarm-agent network, from the perspectives of basic efficiency and adaptability, a performance evaluation system was systematically constructed, comprising the collaborative task completion rate, closed-loop response latency, swarm energy efficiency, collaborative reconfiguration timeliness, and performance retention rate, as well as open-scenario generalization capability. Furthermore, the trade-offs among these metrics were clarified for practical network design and optimization. Moreover, the key challenges were identified in five aspects: security and trustworthiness, heterogeneous agent interoperability, network scalability, real-time control stability, and standardization. Finally, using air-ground embodied agent collaborative tracking and embodied swarm-agent collaborative navigation as typical application scenarios, preliminary verification of the feasibility of two key technologies was provided, such as bio-inspired air-ground low-latency communication and information bottleneck-based embodied agent semantic communication, in the embodied swarm-agent network. This paper aims to provide a systematic reference framework for both theoretical research and industrial practice of embodied swarm-agent networks.关键词：embodied swarm-agent system;embodied swarm-agent network;key characteristics;performance evaluation metrics10|15|0更新时间：2026-09-24
- 摘要：Wireless power transfer technology is considered to have a low-cost advantage in solving the power supply problem of unmanned aerial vehicles (UAVs). However, when the two were combined, the joint optimization problem of power supply incentive and file offloading still needed to be addressed. Based on the interaction relationship between the utilities of UAVs and wireless energy transmission devices, a joint optimization algorithm for UAV offloading optimization and power supply incentive based on the Stackelberg game model was proposed. First, the utility interaction relationship between UAVs and wireless energy transmission devices was modeled using the Stackelberg game model. Subsequently, the model was simplified by deriving it through backward induction. Then, the offloading optimization problem and the power supply incentive problem were solved separately: the transmit power and power supply incentive subproblem was solved by establishing a wireless energy device selection model and an iterative convex optimization threshold transformation, and the file offloading subproblem was solved through problem simplification and derivation. Simulation results showed that, compared with existing algorithms, the proposed algorithm had advantages in reducing UAV energy consumption and power supply expenditure.关键词：unmanned aerial vehicle;task offloading;wireless charging;game theory;convex optimization4|7|0更新时间：2026-09-24
- 摘要：The range ambiguity problem in spaceborne synthetic aperture radar (SAR) observations was addressed. Due to the long observation distance, the echo data received at any instant not only contained the reflected echo of the current pulse but also might be mixed with electromagnetic signals reflected from the previous several pulses, thereby causing range ambiguity. To solve the range ambiguity problem in spaceborne SAR observations, sparse reconstruction and bilinear chirp technology were combined, and cyclic iterative processing was utilized to achieve range ambiguity suppression of the echo signals. In this method, the transmitted signal was first subjected to bilinear modulation, and then a cyclic iterative algorithm was adopted to perform sparse reconstruction and decomposition of the signals in the ambiguity region, thereby achieving ambiguity clutter suppression and high-precision focused imaging of the mixed echo data. The effectiveness and robustness of the method were verified by point target simulation experiments.关键词：synthetic aperture radar;range ambiguity suppression;sparse reconstruction;bilinear frequency modulation6|7|0更新时间：2026-09-24
- 摘要：Abstract: Recently, millimeter-wave communication has attracted extensive research interests for its advantages of large bandwidth and low latency. Meanwhile, age of information (AoI) has been proposed as a key performance indicator reflecting the real-time performance of data at the destination node. However, in millimeter-wave networks, there are still few studies on AoI optimization mechanisms for multicast scheduling. Therefore, the multicast scheduling problem of optimizing AoI in millimeter-wave networks was mainly studied, aiming to minimize the broadcast age of information (BAoI). Firstly, a system model including multiple base stations, multiple user devices and multicast groups was constructed, and the problem of minimizing BAoI through directional selection in millimeter-wave communication and collaborative scheduling across multiple base stations was formulated. Secondly, for the high-dimensional nonlinear optimization problem of beam parameter optimization, a meta-heuristic multicast scheduling algorithm based on differential evolution (DE) was proposed, which integrates adaptive mutation strategies to enhance the global search ability and convergence efficiency in high-dimensional nonlinear search spaces. Simulation experiments show that the proposed algorithm can significantly reduce BAoI under different system parameter settings, verifying its effectiveness in optimizing BAoI in multi-base station millimeter-wave networks.关键词：multi-base stations;millimeter-wave network;age of information;multicast scheduling;differential evolution87|64|0更新时间：2026-09-24
- 摘要：In recent years, deep learning technologies have been widely adopted in the channel state information (CSI)-based fingerprinting localization field, demonstrating high localization accuracy. However, deep learning-based fingerprinting localization models typically rely on multi-layered network structures to extract distinctive location features, a process accompanied by a large number of model parameters and intensive computational operations, which occupy substantial hardware resources on devices. This can be a significant burden, especially for resource-constrained smartphones. To this end,a novel lightweight fingerprinting localization model based on smartphone CSI was proposed. First, a concise and efficient feature extraction module was designed, employing linear and convolutional layers with a minimal number of parameters and incorporating a feature fusion mechanism to achieve both lightweight deployment and strong feature discrimination capability. Second, a data augmentation module was introduced to generate CSI data for non-reference points, thereby expanding the training sample space and significantly enhancing the model's discrimination ability for untrained locations. Experiments were conducted in two typical indoor scenarios. The results show that compared to the most advanced benchmark models, the proposed localization model reduced the average root-mean-square error(RMSE) by 12.5% in two indoor scenarios, and the inference time for a single localization is only 0.08 s. The proposed localization model not only effectively improves positioning accuracy but also significantly reduces the required positioning time.关键词：smartphone channel state information;indoor localization;deep learning;lightweight model100|100|0更新时间：2026-09-24
- 摘要：Semantic segmentation is key in remote sensing, but traditional unimodal methods fail to capture diverse object features fully. Most multimodal studies were confined to bimodal inputs, limiting the exploration of potential complementarity between multimodal data. To address this, an efficient remote sensing image segmentation network with modality expansion capability was proposed. First, a pre-extraction attention head module was designed to distinguish the importance of different modalities in the fusion stage. Second, a cross-modal alignment multi-head attention module was designed to correct offsets and effectively capture heterogeneous features of each modality. Third, a dimensional collaborative linear attention module was designed to enhance the model's global information modeling capability. Finally, an auxiliary loss function was introduced to supervise intermediate feature learning in the decoder stage. The design employs an early fusion strategy, supporting inputs from more than two modalities, and when extended to multiple modalities, the computational cost and parameter overhead increase only slightly. Experiments on the Potsdam and Vaihingen datasets demonstrate that the proposed method outperforms several compared mainstream unimodal and multimodal remote sensing segmentation networks.关键词：semantic segmentation;multimodal fusion;remote sensing images;attention mechanism;modal expansion81|35|0更新时间：2026-09-24
-
Distributed parallel inference of edge large language models: technical applications and challenges AI导读
摘要：Large language model (LLM) have demonstrated immense value across diverse application scenarios, driven by their exceptional cross-domain knowledge generalization and robust capabilities in language understanding and generation. However, in the edge scenarios characterized by strict requirements for low latency and high data privacy, such as autonomous driving and human-computer interaction, traditional centralized cloud-based inference paradigms encounter fundamental challenges. Consequently, deploying LLM within edge environments has emerged as a critical trend. In this paper, distributed parallel technology served as a pivotal enabler for edge LLM deployment, acting as a core mechanism to overcome the computational and memory bottlenecks inherent in edge devices. Focusing on distributed parallel inference for edge LLM, a systematic analysis across three dimensions was conducted, including technical methodologies, deployment practices, and key challenges. The necessity and feasibility of parallel technologies within edge environments were elucidated and a detailed comparative analysis of the efficiency characteristics of various parallel strategies was provided. Furthermore, incorporating insights from actual edge deployments, associated challenges were dissected and optimization directions were identified. This paper aims to provide theoretical references and technical pathways for achieving efficient inference in edge LLM.关键词：edge computing;LLM;distributed parallel technology;Model inference83|125|0更新时间：2026-09-24 - 摘要：To address the challenges of dense terminal access and high computational pressure in edge computing, a secure edge computing offloading architecture based on blockchain and unmanned aerial vehicles (UAVs) was proposed. The architecture leveraged blockchain technology to integrate multiple mobile edge computing (MEC) clusters, enabling trustworthy collaboration of computing resources while enhancing the security of information exchange between UAVs and terminal devices. Firstly, to improve efficiency, this architecture optimized the adopted practical Byzantine fault tolerance consensus algorithm, aiming to reduce the computational energy consumption and latency of blockchain tasks. Then, considering the terminal offloading strategy, UAV trajectories, and the allocation of computing resources for UAVs in both terminal offloading tasks and blockchain tasks, a joint optimization problem was formulated with the goal of minimizing the overall network latency and energy consumption. Finally, since this problem was a mixed-integer nonlinear programming problem, it was solved by alternately using block coordinate descent (BCD) and successive convex approximation (SCA) algorithms. Simulation results show that the proposed scheme significantly reduces the system latency and the energy consumption while ensuring communication security and outperforms the other baseline algorithms.关键词：UAV;MEC;blockchain;privacy protection;trajectory optimization5|6|0更新时间：2026-09-24
- 摘要：With the rapid development of the Internet of things (IoT), massive numbers of heterogeneous devices are continuously being integrated into IoT systems, making the associated security risks the critical issue that demands immediate attention. This is particularly true for resource-constrained embedded devices, for which traditional security solutions that are often reliant on substantial computational resources are unsuitable. To address these challenges, a security attestation scheme for swarm devices in embedded environments was proposed in this paper. A device-initiated attestation mechanism was employed in the scheme to effectively prevent denial of service (DoS) attacks. During the attestation process, control flow attestation was used to ensure runtime security of embedded devices. Blockchain technology was introduced to store large volumes of attestation reports, thereby alleviating the storage burden on embedded devices while enhancing the tamper resistance against the potential alterations of the verifier′s database. A multi-period weighting method was applied to evaluate device trustworthiness, enabling more accurate judgment of device credibility. Furthermore, an online detection mechanism based on a heartbeat protocol was integrated to identify device unavailability caused by physical attacks. Finally, security analysis and performance evaluation demonstrate that the proposed scheme is both secure and suitable for resource-constrained embedded devices.关键词：trusted computing;remote attestation;swarm attestation;control-flow attestation;blockchain technology84|47|0更新时间：2026-09-24
- 摘要：With the rapid evolution of smart cabin interaction technology, contactless gesture recognition has become an essential component of in-cabin human-computer interaction. Traditional methods often face challenges like privacy concerns, high deployment costs, and limited coverage. To overcome these limitations, a gesture mute control system based on vehicle-mounted distributed acoustic devices was proposed. Vehicle-mounted distributed loudspeakers and microphone arrays were reused by the system to construct a low-cost near-field gesture recognition framework. High‑frequency acoustic signals were transmitted, and variations in channel impulse responses induced by hand movements were extracted to realize mute‑gesture sensing. Specifically, an electrical loop channel was first established to obtain reference signals. System delays were dynamically calculated, and time alignment of channel impulse responses was achieved to eliminate feature distortions caused by system latency. Furthermore, a mechanism integrating single‑channel multi‑dimensional feature decision‑making and multi‑channel collaborative decision‑making was presented. Precise discrimination between target gestures and environmental interferences was realized through two-level joint screening. Real-vehicle tests under multiple working conditions demonstrated that a mute-gesture recognition accuracy of over 98%, a false alarm rate lower than 1.5% and a response time shorter than 0.05 s were achieved by the system. Real-time interaction requirements of intelligent cockpits could be satisfied.关键词：in-cabin acoustic sensing;gesture mute control;contactless gesture recognition;smart cabin107|48|0更新时间：2026-09-24
- 摘要：Technologies such as ambient Internet of things (AIoT), low power wide area (LPWA), and low power wake up signal and radio (LP-WUS/WUR) are key enablers for low power system design. For 6th generation mobile networks (6G) low power systems, it is necessary to design a waveform technology that supports low power consumption and wide coverage. While waveform technologies based on amplitude shift keying (ASK) and phase shift keying (PSK) are easy to implement, they cannot meet the wide coverage demands of 6G due to their susceptibility to interference. Conversely, orthogonal frequency division multiplexing (OFDM) waveform technology suffers from high implementation complexity and power consumption. Therefore, this paper focuses on the unified waveform design for 6G low-power systems based on Chirp waveforms, including the principles, performance analysis, hardware architecture design, and 6G standardization analysis. Finally, the future research directions and technical challenges associated with Chirp waveform technology were discussed.关键词：6G;cellular network;chirp waveform;chirp spread spectrum;standardization research122|61|0更新时间：2026-09-24
- 摘要：To address the issues of coverage holes, uneven node distribution, and low energy efficiency caused by random node deployment in wireless sensor networks (WSN), a multi-strategy snake optimizer (MSSO) algorlthm was proposed. Firstly, Tent chaotic mapping and an improved opposition-based learning strategy were employed to generate a high-quality initial population. Secondly, a random-elite alternating guidance mechanism was designed to enhance the exploration capability of the algorithm. Subsequently, adaptlve quantum competitive entanglement and genetic epigenetic regulation mechanisms were introduced during the exploitation phase to increase population diversity and improve convergence. Finally, a dynamic fitness ranking strategy was ada pted to adaptively adjust the selection pressure. Simulation results demonstrated that, when compared with three types of improved algorithms, MSSO exhibited superior performance in terms of both network coverage rate and node distribution uniformity, thereby providing an effective solution for WSN coverage optimization.关键词：wireless sensor network;coverage optimization;snake optimization algorithm;Tent chaotic mapping;elite guidance;adaptive strategy76|35|0更新时间：2026-09-24
- 摘要：Large models are undergoing continuous technological breakthroughs and widespread adoption, with data serving as the foundation throughout their entire lifecycle. However, this pervasive reliance on data exposes such models to significant security and privacy risks and vulnerabilities, making protective strategies and enabling technologies a focal point of current research. Firstly, the data security challenges, privacy risks, and potential attacks were explored in this paper across four critical stages: data collection, model training, model deployment, and model application. Then, the key technologies for data security and privacy preservation were analyzed, including data sanitization, homomorphic encryption, secure multi-party computation, identity authentication and access control, digital watermarking, and differential privacy. Finally, corresponding defense strategies against attacks across the "training-inference-interaction" pipeline were discussed and promising directions for future research were outlined.关键词：Large Models;data security;privacy protection;Attack Defense170|256|0更新时间：2026-09-24
- 摘要：Collaborative tracking of target nodes within the Internet of Underwater Things (IoUT) is a crucial component for achieving continuous coverage and data acquisition in underwater sensing networks. Addressing challenges in multi-AUV collaborative tracking of target nodes, such as high-dimensional observation redundancy, difficulties in identifying critical dynamic features, and insufficient stability in multi-agent collaborative decision-making, this paper proposes the CAM-MADDPG algorithm based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) framework. The proposed method leverages the PC algorithm to construct a directed acyclic causal graph among states, actions, and rewards. Based on the causal connectivity of feature nodes, it generates an attention weight mask to achieve adaptive focusing on critical observational features. Furthermore, it introduces a KL divergence constraint and a counterfactual intervention mechanism to ensure stable weight updates. On this basis, we design a multi-level composite reward function and construct a multi-AUV cooperative node tracking decision-making framework tailored for underwater Internet of Things (UIoT) scenarios. The decision-making performance of the proposed approach is evaluated under an idealized condition where collaborative information among agents is assumed to be fully available. Simulation results demonstrate that CAM-MADDPG outperforms baseline methods such as MADDPG, MAAC, MASAC, and MATD3 in metrics including tracking accuracy, velocity consistency, and return convergence stability, thereby validating the effectiveness and robustness of the proposed method.关键词：multi-agent reinforcement learning;MADDPG;attention mechanism;Autonomous underwater vehicle;Collaborative target tracking;Markov decision process7|2|0更新时间：2026-09-22
- 摘要：Device-to-Device (D2D) communication is confronted with critical security threats arising from its inherently short link distances. To address this, the Active Reconfigurable Intelligent Surface (ARIS) is employed to actively amplify reflected signals so as to reinforce the legitimate link, while simultaneously leveraging the near-field spherical wavefront characteristic to suppress eavesdropping directions. In this paper, we investigate an ARIS-assisted D2D secure transmission system operating in near-field channels and formulate a security energy efficiency maximization problem. Given the high coupling among variables in the objective function and constraints, an Alternating Optimization (AO) based algorithm is proposed to decouple the original problem into several subproblems. To tackle the non-convexity, we successively adopt Successive Convex Approximation (SCA) and Semidefinite Relaxation (SDR) to iteratively solve each subproblem. Simulation results verify that the proposed scheme significantly enhances the system security energy efficiency, thereby corroborating the synergistic benefits of active amplification and coherent phase control of ARIS in near-field D2D secure communications.关键词：active reconfigurable intelligent surface (ARIS);device-to-device (D2D);near field communication;secure communication3|0|0更新时间：2026-08-31
- 摘要：Sixth generation mobile communications (6G) is driving the evolution of wireless networks from traditional information transmission toward integrated systems that combine environmental perception with spatial intelligence. As the central medium through which 6G networks interact with the physical environment, electromagnetic space and its precise cognition directly determines the synergistic efficiency of perception, communication, and decision making; thus, it serves as a critical prerequisite for realizing on demand services in 6G networks. However, the inherent complexity of electromagnetic space makes it challenging to achieve high precision, real-time, and predictable digital representations, thereby constraining the development of intelligent digital twin technologies for 6G electromagnetic environments. Traditional modeling approaches struggle to simultaneously balance accuracy, real-time performance, and generalizability; moreover, existing research often focuses on only one of the three fundamental elements-radiation sources (“Source”), propagation environments (“Environment”), and electromagnetic fields (“Field”)-without systematically characterizing their coupled relationships. To address these issues, an integrated “Source-Environment-Field” cognitive model for intelligent electromagnetic space twins was proposed. The model achieves a fine-grained representation of the “Source” through techniques such as parameter estimation and behavior recognition, completes the physical reconstruction of the “Environment” through 3D geometric reconstruction and electromagnetic propagation-effect modeling, and performs digital-twin modeling of the “Field” using multiscale strategies and physics-data fusion methods. Furthermore, from an integrated cognitive perspective, this paper summarizes the roles of existing techniques in three collaborative links: existing techniques such as differentiable ray tracing, neural radiance fields, and generative radio map (RM) can mainly support the “Source-Environment-to-Field” forward prediction; source localization and parameter inversion methods can support the “Field-Environment-to-Source” inverse solving; and environmental structure and electromagnetic material inversion methods can support the “Field-Source-to-Environment” reverse inference. Subsequently, the paper explores the potential applications of intelligent electromagnetic space twins across various typical scenarios, including 6G networks, integrated space-air-ground communication systems, emergency rescue operations, and national defense security. Finally, the paper summarizes the major challenges currently facing this field of research and outlines future directions for the development of unified electromagnetic space modeling theories, physics-constrained da-ta-driven methodologies, and collaborative evolutionary approaches involving large-scale models-all geared toward the 6G era.关键词：6G;electromagnetic intelligence twin;Source-Environment-Field;digital representation;electromagnetic space;AI empowerment6|8|0更新时间：2026-08-10
- 摘要：Considering the issues of communication and sensing securities for the reconfigurable intelligent surface (RIS) enabled integrated sensing and communication system, a joint design for the transmit covariance matrix and the reflection matrix of the RIS is proposed to maximize the ergodic secrecy rate under the constraints of legal sensing signal-to-noise ratio (SNR), eavesdropper sensing SNR, transmit power, and the constant modulus for the reflection matrix of the RIS. Firstly, the asymptotic ergodic secrecy rate lower bound and the asymptotic sensing SNRs are derived by exploiting the large dimensional random matrix theory. Then, based on the statistical channel state information, the optimal closed-form optimal solution for the transmit covariance matrix is derived by exploiting the modified water-filling algorithm and the first-order Taylor series expansion, and sub-optimal the reflection matrix of the RIS is obtained by employing the barrier method and the projected gradient ascent method. Finally, numerical simulations validate the accuracy of the large-system analysis and the superiority of the proposed method.关键词：integrated sensing and communication (ISAC);reconfigurable intelligent surface (RIS);statistical channel state information (CSI);physical layer security238|75|0更新时间：2026-08-10
- 摘要：To address issues in class-incremental object detection models such as noise interference during detection, instability of old class boundaries, and imbalance between learning old and new classes, this paper proposes improvements to the CL-DETR model and introduces a class-incremental object detection model CC-DETR based on curriculum learning distillation. First, a noise-resistant collaborative purification module is constructed to filter label noise and suppress background noise interference while performing pseudo-label purification and foreground-guided distillation. Secondly, a boundary distillation method based on knowledge consolidation is introduced, applying KL distillation only to old class channels to stabilize the decision boundaries of old classes. Finally, we designed a curriculum-style learning distillation strategy, enabling the model to quickly learn new classes in the early stages of training, and dynamically adjust the distillation weights as training progresses to stabilize old classes while learning new ones. Experiments show that under the 70+10 and 40+40 splits of the COCO dataset, detection accuracy reached 40.9% and 39.1%, respectively, improving over CL-DETR by 1.9% and 2.4%, and outperforming current mainstream class-incremental object detection algorithms. Meanwhile, with the same inference structure, FPS increased by 9.7% and training time decreased by 25.4%, making it better suited for practical class-incremental object detection scenarios.关键词：Object detection;Class incremental;Knowledge distillation;Dynamic weight guided59|9|0更新时间：2026-08-10
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
- Total Visits：129566 Online Users：6
0