由于飞行器使用的低可观察性设计,为了提高雷达高分辨率距离像(HRRP)目标识别的鲁棒性,提出了一种基于全局目标概率HRRP 目标识别算法,该算法采用深度卷积神经网络从HRRP 自动提取目标特征并进行目标识别。提出利用多基地雷达系统提供的空间分集,对不同目标类别的分类器的本地输出概率进行平均,将最高的全局目标概率与阈值进行比较,以获得用于最终分类的全局目标概率,从而提高了目标识别的鲁棒性,使隐形目标的检测和识别更加可靠。仿真结果表明,即使在低信噪比的情况下,所提出的算法可以显著提高分类的可靠性。

Due to the low observability design of the aircraft, in order to improve the robustness of radar high resolution range profile (HRRP) target recognition, the HRRP target recognition algorithm based on global target probability is proposed in this paper, which uses deep convolutional neural network automatically extracts target features from HRRP and performs target recognition. It is proposed to use the spatial diversity provided by the multistatic radar system to average the local output probabilities of the classifiers of different target categories, and compare the highest global target probability with a threshold to obtain the global target probability for the final classification. It improves the robustness of target recognition, and makes the detection and recognition of invisible targets more reliable. Simulation results show that even in the case of low signal-to-noise ratio, the proposed algorithm can significantly improve the reliability of classification.

TN959

国家自然科学基金资助项目(61801435);江苏高校自然科学研究面上项目(16KJB510008);江苏高校骨干专业建设工程基金资助项目(苏教高[2017] 17号);2018 年江苏省“333”工程第三层次培养对象资助项目(苏才人办[2018]6 号);江苏省高校产教融合集成平台项目(苏教职[2019]31 号)