Chunhui Qu

Synthetic Aperture Radar (SAR) supports all-weather, day-and-night Earth observation, making it vital for military and civilian applications. While deep learning has advanced SAR target detection, state-of-the-art (SOTA) methods often directly adopt optical data augmentations, such as rotation, scaling, and Mixup, which violate SAR imaging physics and lead to artifacts, resolution distortion, and physically implausible scattering patterns. This work reveals that such physics-violating augmentations consistently compromise the detection reliability of real-time YOLO-series detectors, with the most severe drops occurring under strict localization criteria corresponding to higher IoU thresholds. To address this issue, we propose physics-preserving augmentations (PPA) that respect SAR domain priors. Extensive experiments on three representative datasets (SAR-AIRcraft-1.0, SIVED, and SRSDD) demonstrate the consistent superiority of PPA over optical augmentations in the context of SAR target detection. Notably, PPA enables the lightweight YOLOv11n to achieve 79.2% mAP75 on SAR-AIRcraft-1.0, comparable to or superior to existing SOTA SAR detectors while retaining real-time efficiency.

The development of intelligent transportation systems (ITS) has attracted significant attention to connected and autonomous vehicles (CAVs). It is urgent to investigate multi-CAV intelligent cruise control solutions in mixed traffic environments. In addition, the impact of platoon dynamics and time delays, induced by shared wireless communications, data processing, and actuation cannot be ignored. This article investigates the development of a multi-agent deep reinforcement learning (MADRL) controller tailored for CAVs operating within mixed and dynamic traffic scenarios that involve time delays. Firstly, the error dynamics in the discrete-time domain for each subplatoon is derived by considering the time-varying delays and leading vehicle states, and then the optimal CAV cruise control problem is formulated. Subsequently, the partially observable Markov game (POMG) is used to construct the multi-agent environment, and then a centralized training decentralized execution (CTDE) algorithm framework is proposed based on the multi-agent deep deterministic policy gradient (MADDPG) method. Finally, the computational complexity and the influence of delay are analyzed. The simulation results illustrate the effectiveness of the proposed intelligent algorithm.

Power converters become increasingly important in the power industry. To enhance design reliability and efficiency, real-time simulation is a popular method for testing control units of these devices. FPGA-based real-time simulation is considered a genuine alternative to CPU-based applications as control unit frequencies increase. Although automated tools can map simulation methods from floating-point high-level languages to fixed-point FPGA, manually selecting the fixed-point word length to meet precision requirements involves a significant amount of tedious and repetitive work. Word length selection also impacts FPGA hardware resource usage. This paper proposes an adaptive word length selection method under accuracy constraints to minimize hardware resources. The innovation lies in using error noise models to compute analytical expressions relating simulation model accuracy to word length. The adaptive word length selection process is combined with commercial high-level synthesis tools to achieve high resource optimization performance. A three-phase inverter example demonstrates the benefits of our method in calculating model accuracy and optimizing FPGA hardware resources in simulations.

Field-programmable gate array (FPGA)-based real-time simulation serves as a powerful tool for testing and validating power electronics system (PES). However, accurate real-time simulation in high-frequency applications demands extremely small time steps, which poses significant computational and memory challenges. Explicit methods can achieve small time steps due to low-computational burdens, but have stability issues. In this article, the authors propose a topology-aware matrix partitioning (TA-MP) method, which adopts the implicit numerical algorithm to ensure stability, and uses iterative methods to solve the implicit equations. In detail, the PES matrix is partitioned into blocks with clear topological meanings, and its constant diagonal blocks are used to construct a constant iterative matrix. Since the iterative matrix is constant, the TA-MP method bypasses the limitations of using iterative methods in real-time simulations by predetermining the number of iterations. The TA-MP method eliminates the need to store inverse matrices for different topologies, requiring only matrix-vector multiplications, which are efficient on FPGA. The TA-MP method's effectiveness is validated with n-port active bridge converters on an FPGA board with a 25 ns step size, ensuring simulation accuracy at the highest switching frequency of 200 kHz while consuming only 1/15 the memory resources of the conventional method.