培训(气象学)
对抗制
计算机科学
雷达
雷达跟踪器
人工智能
雷达成像
便携式雷达
工程类
钥匙(锁)
训练集
空中交通管制雷达信标系统
雷达工程细节
雷达锁定
三维雷达
合成孔径雷达
计算机视觉
双基地雷达
雷达探测
探地雷达
实时计算
遥感
雷达配置和类型
自动目标识别
杂乱
鉴定(生物学)
天基雷达
雷达系统
模拟
遥控水下航行器
作者
Ke Wang,Xiaoyi Pan,Xiaofeng Ai,Zhiming Xu,Feng Zhao,Shunping Xiao,Siwei Chen,Yafei Song
标识
DOI:10.1109/taes.2026.3685957
摘要
The electromagnetic space of modern battlefields is becoming increasingly complex. High-Resolution Range Profile (HRRP) has emerged as a core data carrier in the field of Radar Automatic Target Recognition (RATR), due to its advantages of easy acquisition, low storage resource requirements, and high processing efficiency. However, the complex electromagnetic environment inten sifies HRRP's sensitivity to noise and target attitude changes. Existing HRRP recognition methods generally suffer from weak feature extraction capabilities, poor dynamic adaptability, and insufficient robustness in complex environments. To address the aforementioned issues, this paper proposes an Adv-STMamba model, which is built upon a Spatial-Temporal Mamba module (STMamba) and integrates with to dedicated modules: Adversarial Label Smoothing and Adaptive Attention Gate modules. Specifically, Adversarial Label Smoothing resolves the problem of model's overconfidence in labels by generating adversarial soft labels that maximize model loss; STMamba enables in-depth mining of spatiotemporal features in HRRP sequences; Adaptive Attention Gate module achieves efficient spatiotemporal feature fusion by adaptively assigning feature weights, thereby enhancing model's robustness against attitude disturbances. To ensure the persuasiveness of experimental data, this paper converts the public MSTAR dataset into an HRRP dataset and conducts verification through comparative experiments, robustness experiments, complexity experiments, and ablation experiments. The experimental results show that Adv-STMamba model can effectively extract deep-seated features from HRRP data under different signal-to-noise ratio (SNR) scenarios. It achieves a recognition accuracy of 99.67% under the high SNR scenario of 30dB and maintains a recognition accuracy of 97.30% even under the low SNR scenario of 0dB. This fully verifies the effectiveness of Adv-STMamba model in HRRP recognition tasks under complex environments.
科研通智能强力驱动
Strongly Powered by AbleSci AI