Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Junyi Wu has a total of 24 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A Multi-Source Data Fusion and PSO-Transformer Decision Support System for Gold Price Volatility Forecasting and Risk Warning

    by Xueping Diao, Xueqiang Lv, Weizheng Wang, Junyi Wu

    Gold volatility responds to shocks and sentiment, yet the two are rarely fused, and forecasts seldom become actionable. We link a multi-source fusion subsystem to a risk-warning layer through a Transformer whose three pre-registered hyperparameters are searched by particle swarm optimization (PSO-Transformer). It fuses video sentiment, headline scoring and market factors screened to 20 of 21 candidates; the decision side maps the predicted realized variance RVt/Mt (bp2, non-annualized) into three traffic-light tiers of a calibrated predictive distribution. On 5-min spot gold data (January–June 2025, 126 trading days), the PSO-Transformer attains a one-step MAE of 8.038, a 36.5% average improvement over seven baselines under protocol P1, where selection and evaluation share one window; under the leak-free protocol P2, the same pipeline attains 10.16 against 8.82 for the GRU, so the first place reflects the P1 design, not an independent out-of-sample comparison. Diebold–Mariano gains over GARCH, SVM and XGBoost survive a Holm correction. Walk-forward backtesting over 59 origins with persistence-anchored alerts yields 67.8% tier accuracy, 78.8% high-risk precision and 92.9% recall, with the neural engine fed into the same rule attaining 45.5% against 61.4%. One instrument, six months: a documented architecture, not a demonstrated transfer.

  • Open Access

    MulRobBench: A Decision-Level Benchmark for Safe and Security-Policy-Compliant Multimodal UAV Agents

    by Belal S. Alsinglawi, Weizheng Wang, Junyi Wu, Yi Jiang, Lianhai Lin, Merouane Debbah, Izzat Alsmadi

    Smart-city airspace is transforming Uncrewed Aerial Vehicles (UAVs) from passive sensing platforms into cyber-physical decision makers that must follow operational rules under degraded observations and ambiguous language. Existing UAV and multimodal benchmarks evaluate perception, navigation, collaboration, and reasoning, but few assess whether physical evidence, protocol constraints, and action risk remain coupled during critical decisions. We introduce MulRobBench, an offline, protocol-conditioned benchmark for Vision-Language-Action (VLA) UAV agents in smart-city environments. MulRobBench integrates real UAV multimodal observations, protocol-level security policies, and action-level cyber-physical safety into a unified evaluation framework. The benchmark contains 3,024 samples spanning 17 task taxonomy nodes and 12 scoring dimensions across four stages: operational context understanding, multimodal evidence arbitration, degradation-aware reasoning, and risk-aware action planning. Evaluation combines semantic scoring with structural diagnostics, including policy compliance, format compliance, unsafe actions, parsing failures, and dimension-level validity. Across 17 multimodal models, the best semantic protocol-decision score reaches only 0.5141, while the best strict mean scoring-dimension accuracy is 0.1599. A controlled 20-anchor modality-ablation study changes 4-15 action selections per model, confirming that both visual and textual inputs influence decisions. Analysis identifies modality-trust selection, constraint extraction, glare, missing data, and operator shorthand as the primary causes of decision instability. MulRobBench provides a reproducible benchmark for trustworthy multimodal UAV decision making under realistic operational constraints.

  • Lightweight Feature-Enhanced Convolutional Module via Channel Decoupling and Dual-Pooling Attention

    by Yong Niu, Yi Jiang, Junyi Wu, Jun Ma, Ze Li

    Lightweight attention-enhanced convolutions have been widely adopted to reduce computational costs in visual models, but most existing methods suffer from coarse channel grouping, insufficient functional decoupling, and lack of dynamic weighting mechanisms. These limitations lead to redundant feature responses, weak spatial modeling, and degraded representation capability under low-computation constraints. To address these drawbacks, this paper proposes a channel-decoupled dual-pooling attention convolution (CDDPConv) for efficient lightweight feature extraction. The proposed module retains the low-cost advantages of partial convolution while explicitly implementing multi-branch functional decoupling along the channel dimension. Three independent Q, K, and V branches are constructed to separately capture local salient features, global contextual information, and intact spatial details via max pooling, average pooling, and 3×3 convolution. Combined with dynamic weighted spatial attention and residual connections, CD-DPConv adaptively recalibrates feature responses and enhances cross-branch information interaction. Comparative experiments show that the proposed method outperforms state-of-the-art lightweight attention convolutions in comprehensive performance under identical computational budgets. The integration of functional decoupling and dynamic weighting effectively balances model compactness and high-precision feature learning, providing a reliable plug-and-play solution for general lightweight visual networks.

  • UAV-enabled Maritime Search and Rescue: A Human Detection Perspective

    by Junyi Wu, Yi Jiang, Yuxia Gao, Liqin Tian, Yong Niu

    Maritime Search and Rescue (SAR) is critical for ensuring life safety at sea. Uncrewed Aerial Vehicles (UAVs), with their high mobility and flexible viewing angles, have emerged as a pivotal technology for locating persons–in–water. However, human detection in ocean faces significant challenges due to the small scale of targets and heavy interference from complex backgrounds such as sea waves, sun glitters, and buoys, which complicates feature extraction and leads to high rates of missed detections. To address these challenges, this paper proposes TDA–Net, an efficient feature–enhanced detection model based on an improved DEIM framework. We introduce a tri– path feature–level fusion module to effectively integrate multi– modal image data, thereby enhancing the model‘s representation capability. To overcome the scale-insensitivity of traditional loss functions toward small maritime targets, we propose Scale-Enhanced Matchability-Aware Loss (SE–MAL) to reconstruct objective function. Experimental results demonstrate that TDA– Net achieves an mAP50 of 84.2% on the maritime human detection dataset, marking a 1.5 improvement over the baseline. Simultaneously, by integrating efficient Depthwise Separable Convolution (DWConv) and Fast Fourier Transform (FFT) filtering, the model reduces computational complexity by 0.2 GFLOPs while introducing only a marginal increase in parameters. This confirms its superior performance in balancing high precision with high efficiency for practical maritime UAV SAR missions.

  • Open Access

    A lightweight insulator defect detection algorithm based on drone images for power line inspection

    by Junyi Wu, Weizheng Wang, Wanyan Du, Zhi Liu, Lianhai Lin, Yuchen Cui, Chong Wang, Liqin Tian

    Accompanying the enhancement of unattended systems based on Unmanned Aerial Vehicles (UAVs) and UAV base station, the lightweight model deployment of unattended systems is of great significance for inspecting the condition of transmission power lines and ensuring the safe and stable operation of the power grid. Current insulator defect detection methods are tasked with small–object detection in complex environments, with their accuracy often improved by optimizing network structures only. Considering both parameter and computational efficiency, we developed LFL–YOLO (LSDHead–FINs–LAMP–YOLOv8n) that is suitable for edge computing. Before the spatial pyramid transmits feature maps to the detection head, some levels are skipped based on inter–level sparsity, and the model’s ability to perceive scale and task-aware attention for complex backgrounds and small objects is dynamically enhanced. By combining this with spatial attention, a high–performance LSDHead is proposed. Secondly, a lightweight FINs is proposed to improve the neck of YOLOv8. This structure ensures effective interaction and feature penetration across different channels of local features. Finally, layer–adaptive unstructured pruning based on magnitude is performed. The pruned model not only reduces computational complexity and model size, significantly improving the model’s real–time performance, but also enhances detection accuracy. Lastly, to align more closely with real–world engineering applications, this study uses public datasets to verify the model’s generalization effect and validates the practical application effect on a dataset. Experimental results show that compared to the current benchmark, LFL–YOLO achieves significant improvements in accuracy and inference speed. Specifically, LFL–YOLO achieves an accuracy of 97.4%, compared to YOLOv8n, reduces the number of parameters by 23.5% and decreases computation by 9.6%. The proposed method has practical applicability in obtaining aerial images of transmission lines in complex environments.


  • Open Access

    Multi-objective dung beetle optimization algorithm: A novel algorithm for solving complex multi-objective optimization problems

    by Wenxing Wu, Liqin Tian, Junyi Wu, Lianhai Lin

    Many increasingly complex multi-objective optimization problems are emerging, and there is an urgent need to develop new multi-objective optimization algorithms to meet the challenges. This study introduces the Multi-Objective Dung Beetle Optimization Algorithm (MODBO), which integrates competitive and neighborhood mechanisms to tackle such problems, Thanks to the dung beetle optimization algorithm’s fast convergence and robust optimization finding ability in single-objective optimization algorithms. The introduction of non-dominated sorting allows the Dung Beetle Optimization Algorithm to solve multi-objective optimization problems (MOPs). To make the Dung Beetle Optimization Algorithm maintain good search ability in searching, we introduce a Competition mechanism to guide the particles’ global optimal search and a Neighborhood mechanism to guide the particles’ local optimal value search. An external archive is introduced to make each generation positionally optimal. Finally, to analyze whether the MODBO algorithm’s improved strategy is effective, a comparison with the nine algorithms on CEC2020 was made, and the 3D sensor deployment problem was used to demonstrate that the MODBO algorithm can solve realistic problems.


  • Open Access

    Coverage optimization and node minimization in WSNs: an enhanced hybrid PSO approach with spatial position encoding

    by Yinghua Tong, Lianhai Lin, Liqin Tian, Zhigang Wang, Wenxing Wu, Junyi Wu

    Wireless sensor networks (WSNs) are widely used in various applications requiring efficient coverage and minimal resource utilization. This paper presents an enhanced hybrid particle swarm optimization (EHPSO) algorithm that incorporates a spatial position encoding (SPE) strategy to optimize coverage while dynamically adjusting the number of sensors deployed in WSNs. The proposed approach leverages the strengths of particle swarm optimization (PSO) by integrating it with the SPE mechanism, which effectively guides the search process towards high-quality solutions. The EHPSO algorithm is designed to balance exploration and exploitation capabilities, enabling dynamic node adjustment and ensuring robust performance across different network configurations and environmental conditions. Extensive simulations are conducted to evaluate the performance of the proposed method against state-of-the-art algorithms in terms of coverage quality and node count. A multi-objective optimization model is also established, further illustrating the algorithm’s performance and its effectiveness in balancing the number of sensors and coverage rate. Results demonstrate improvements in coverage optimization and reduction of node deployment compared to existing methods. This research contributes to more efficient and cost-effective deployment strategies for WSNs, particularly in scenarios where resources are limited and optimal coverage is critical.