摘要
在逆合成孔径雷达(Inverse Synthetic Aperture Radar,ISAR)成像领域,运动补偿是确保高质量图像生成的关键环节。包络对齐(Range Alignment,RA)作为运动补偿的首要步骤,对于校正由平动分量引起的回波信号包络偏移至关重要。本文提出了一种基于复数域卷积神经网络(Complex-Valued Convolutional Neural Network,CVCNN)的包络对齐新方法,旨在通过深度学习策略提升包络对齐的精度与计算效率。本文所提方法利用了卷积神经网络强大的特征学习能力,构建了一个能够映射一维距离像与包络补偿量之间复杂关系的模型。通过将传统的实值卷积神经网络拓展至复数域,不仅完整保留了回波信号中的相位信息,而且有效引入了复数域残差块及线性连接机制,进一步精细化了网络结构设计。这种架构改进使得所提算法能实现低信噪比(Signal-to-Noise Ratio,SNR)条件下对ISAR距离像的高效包络对齐。在数据生成方面,本文基于雷达仿真参数,通过成像模拟仿真构建了ISAR回波数据集。该数据集经过归一化处理后,输入网络进行训练,使网络能够学习从未对齐回波到对应补偿量的映射关系。本文所提方法采用迁移学习策略,对基于仿真数据预训练的模型进行微调,以适应实测数据。这一策略不仅增强了结果的可靠性,同时也大幅缩短了模型的迭代周期。在实验验证方面,本文采用仿真与实测数据进行综合测试,以包络对齐精度、成像结果质量和计算效率为评价指标,全面验证了算法的有效性。实验结果表明,在不同信噪比条件下,本文所提方法均展现出了优越的包络对齐性能,进而可以实现高质量成像,同时在计算效率上也具有显著优势。
In the field of Inverse Synthetic Aperture Radar(ISAR)imaging,motion compensation is a critical step to ensure high-quality image generation.Range alignment(RA)serves as the primary step in motion compensation,correcting range offsets in echo signals caused by translational motion.This article introduces a new range alignment method based on a Complex-Valued Convolutional Neural Network(CV-CNN),aiming to enhance the accuracy and computational efficiency of range alignment through deep learning techniques.The proposed method leverages the potent feature-learning capabilities of convolutional neural networks to construct a model that maps the complex relationship between one-dimensional distance profiles and range compensation amounts.By extending traditional real-valued convolutional networks into the complex domain,our approach preserves the phase information in the echo signal and effectively incorporates residual blocks and linear connection mechanisms in the complex domain,refining the network structure design.These architectural improvements allow the proposed algorithm to achieve efficient range alignment of ISAR range profiles,even under low signal-to-noise ratio(SNR)conditions.For data generation,an ISAR echo dataset was constructed through simulation based on relevant radar parameters.After normalization,this dataset was input into the network for training,enabling it to learn the mapping from unaligned echoes to the corresponding compensation quantities.The method employs a transfer learning strategy to fine-tune the pre-trained model on simulation data,adapting it to actual measurement data.This strategy enhances the reliability of the results and significantly shortens the model’s iteration cycle.In terms of experimental verification,the article rigorously evaluates the algorithm’s effectiveness using both simulated and measured data,focusing on range alignment accuracy,imaging result quality,and computational efficiency as key indicators.The experimental results demonstrate that,under various signal-to-noise ratio conditions,the method exhibits exceptional envelope alignment performance,yielding high-quality imaging and substantial advantages in computational efficiency.
作者
王勇
夏浩然
刘明帆
WANG Yong;XIA Haoran;LIU Mingfan(Research Institute of Electronic Engineering Technology,Harbin Institute of Technology,Harbin,Heilongjiang 150001,China)
出处
《信号处理》
北大核心
2025年第3期409-425,共17页
Journal of Signal Processing
基金
国家杰出青年科学基金(62325104)。
关键词
逆合成孔径雷达
包络对齐
复数域卷积神经网络
有监督学习
inverse synthetic aperture radar
range alignment
complex-valued convolutional neural network
super‐vised learning
作者简介
通信作者:王勇,男,1979年生,吉林扶余人。哈尔滨工业大学电子与信息工程学院教授,博士生导师,主要研究方向为雷达成像技术。E-mail:wangyong6012@hit.edu.cn;夏浩然,男,2002年生,辽宁沈阳人。哈尔滨工业大学电子与信息工程学院硕士研究生,主要研究方向为机器学习与雷达成像技术。E-mail:120L020827@stu.hit.edu.cn;刘明帆,男,2001年生,黑龙江哈尔滨人。哈尔滨工业大学电子与信息工程学院硕士研究生,主要研究方向为机器学习在雷达信号处理中的应用。E-mail:liumingfan@stu.hit.edu.cn。
