ABSTRACT The snow ablation optimizer (SAO) is a novel meta‐heuristic algorithm inspired by the natural processes of snow melting and sublimation. SAO employs a dual‐population dynamic adjustment mechanism to balance exploration and exploitation, demonstrating significant performance in global optimization. However, when dealing with complex optimization problems, the SAO algorithm tends to suffer from low convergence accuracy and premature convergence. To address these issues, this paper proposes a multi‐strategy snow ablation optimizer (MSSAO). Firstly, it adopts an external archive mechanism to prevent convergence to local optima. Secondly, it utilizes an adaptive t ‐distribution perturbation strategy to achieve a balance between exploration and exploitation. Finally, it employs a mirror reflection learning strategy to maintain population diversity. Comprehensive experiments conducted on the CEC2017 and CEC2022 test suites demonstrate that MSSAO significantly enhances both global optimization performance and convergence accuracy. Furthermore, its effectiveness in engineering optimization is demonstrated through a DC motor control system.
ABSTRACT Long‐term time series forecasting (LTSF) plays a crucial role in practical applications such as power dispatching, financial market analysis, and climate prediction. To tackle core challenges in LTSF, including the difficulty of capturing nonlinear patterns and insufficient multi‐scale feature extraction, we propose a novel Multi‐scale Frequency Domain Decomposition Network (MFDNet). The proposed model features a hierarchical frequency‐domain processing architecture driven by three key innovations. First, a reversible instance normalization (RevIN) layer is introduced to mitigate distribution shifts in non‐stationary data, thereby enhancing model generalization. Second, a differentiable wavelet decomposition module with learnable, orthogonality‐constrained filters is designed to explicitly separate trend and seasonal components, where symmetric boundary extension is adopted to avoid the discontinuities that zero padding would introduce at the two ends of the observation window. Third, an attention‐guided adaptive frequency selection mechanism dynamically emphasizes critical frequency components across multiple scales and temporal contexts. Experimental results on seven benchmark datasets across four domains demonstrate that MFDNet significantly outperforms or matches state‐of‐the‐art methods. For instance, on the ETTm2 dataset, MFDNet reduces the Mean Squared Error (MSE) by compared to SparseTSF, and on the Traffic dataset, the average MSE decreases by compared to iTransformer, highlighting the model's predictive accuracy and robustness.
ABSTRACT In multi‐node hardware‐in‐the‐loop (HIL) simulation systems, efficient coordination and deterministic data exchange across simulation nodes have become critical challenges for ensuring system stability, and simulation fidelity. Traditional TCP/IP‐based approaches struggle to meet stringent latency constraints due to high protocol overhead. Meanwhile, technologies such as PCIe switch based on NTB and RDMA are hindered by hardware lock‐in, high cost, and lack of transparent access. This paper presents a Smart PCIe Switch ( SmartSW ) that integrates optoelectronic expansion to overcome the distance limitations of electrical PCIe signaling. By implementing programmable switching logic on an FPGA, SmartSW offloads data distribution logic from target hosts to switch nodes, enabling a “single read, on‐demand distribution” packet processing paradigm that breaks through PCIe's native multicast limitations. Furthermore, end‐to‐end deterministic communication is achieved through zero‐copy DMA and interrupt‐free awareness mechanisms. Experimental results show that, when transferring 4 KB of data between two simulation nodes, the SmartSW pre‐splits packets based on routing tables, reducing transmission latency by up to 32%. High‐payload‐ratio data transfers are completed within less than 10 μs, demonstrating a highly efficient, flexible, and hardware‐software co‐designed interconnect solution for large‐scale, real‐time, multi‐node simulation systems.
ABSTRACT As the spectral dimension of hyperspectral images (HSI) increases, reducing spectral redundancy becomes critical for effective HSI classification. Existing approaches are generally categorized as either feature extraction or band selection. Band selection methods more effectively preserve the original data's characteristics; however, traditional enumeration techniques are computationally intensive. Fuzzy clustering, an unsupervised band selection approach, efficiently groups unlabeled data but is sensitive to initialization, while gradient‐based methods frequently converge to local optima. Swarm intelligence methods provide flexibility and robust optimization performance, yet they continue to encounter challenges related to global search capability and local optima. To address these limitations, this study introduces an enhanced hiking optimization algorithm featuring a modified search mode, improved exploitation precision, and local mutation enhancement for high‐dimensional tasks. By integrating this algorithm with fuzzy clustering, an adaptive band selection method, EHOA‐FCM, is proposed to advance fuzzy clustering‐based band selection. Furthermore, the inter‐band measurement formula and band selection strategy are refined to better accommodate HSI characteristics. Experiments conducted on three benchmark hyperspectral datasets demonstrate that the proposed method selects band subsets with more uniform distribution and broader coverage, resulting in significant improvements in HSI classification accuracy.
ABSTRACT Unmanned aerial vehicle (UAV) imagery presents persistent challenges for object detection because targets are often extremely small, weakly textured, densely distributed, and embedded in complex backgrounds. This study proposes a YOLO11s‐based Progressive Dynamic Enhancement Network (PDFN) that coordinates feature extraction, cross‐layer fusion, and detection prediction. The backbone combines Receptive‐Field Modulated C3k2 (RFM‐C3k2) and Multi‐Scale Pyramid Convolution (MSPC), the neck integrates the Dynamic Feature Refinement Module (DFRM) and Scale Alignment Calibration Module (SACM), and the Scale‐Adaptive Decoupled Head (SAD‐Head) introduces classification reweighting and local regression compensation. Under three independent runs with seeds 0, 42, and 123 on VisDrone2019‐DET, PDFN achieves 42.63% ± 0.15% mAP@0.5, 25.80% ± 0.10% mAP@0.5:0.95, and 26.23% ± 0.15% APsmall (mean ± sample standard deviation), exceeding YOLO11s by 3.20, 2.23, and 2.47 percentage points, respectively. Averaged over the final 20 epochs and three runs, PDFN reduces validation box, classification, and distribution focal losses by 4.76%, 6.74%, and 4.12%. Quantitative feature‐response analysis shows that the target‐to‐background response ratio increases from 2.54 to 3.19. PDFN contains 8.9 M parameters, requires 21.8 GFLOPs, and achieves 123 FPS under the stated forward‐pass timing protocol. Together, ablation, partial‐path, multi‐dataset, visualization, and failure‐case analyses provide a comprehensive evaluation of co‐ordinated full‐link enhancement across diverse UAV scenarios.
