| 引用本文: | 张 驰， 安洪阳， 娄明悦， 李中余， 武俊杰， 杨建宇. SAR射频干扰区域⁃强度特征提取与联合评估网络[J]. 雷达科学与技术, 2024, 22(4): 391-399.[点击复制] |
| ZHANG Chi, AN Hongyang, LOU Mingyue, LI Zhongyu, WU Junjie, YANG Jianyu. SAR Radio Frequency Interference⁃Region Intensity Feature Extraction and Joint Evaluation Network[J]. Radar Science and Technology, 2024, 22(4): 391-399.[点击复制] |
|
| 摘要: |
| 由于射频信号的广泛存在，合成孔径雷达（Synthetic Aperture Radar, SAR）在成像的过程中容易受到各类射频干扰（Radio Frequency Interference, RFI）的影响，这会导致获得的SAR图像质量下降，从而对后续的信息提取和目标识别等过程产生很大的影响。因此，衡量SAR图像受射频干扰的影响程度就尤为重要。然而，现有评估方法的鲁棒性通常较低，并且在评估时未考虑SAR图像受RFI影响的区域大小，因此本文提出了干扰区域?强度特征提取与联合评估网络。所提出的网络包含两个模块，干扰强度特征提取模块用于提取输入SAR图像中的干扰强度信息，干扰区域特征提取模块则侧重于干扰区域大小与边界信息的获取。由于SAR图像的尺寸一般比较大，因此本文在干扰强度特征提取模块中采用了多级残差和多层特征融合结构，用于加强模型的特征提取和复用能力；同时在干扰区域特征提取模块中侧重于保留最关键的区域边界特征。此外，本文还建立了SAR受RFI影响的图片数据集用于评估所提出网络的效果。对比实验的结果表明，本文所提出的网络评估结果优于其他现有方法，能够衡量SAR图像受RFI的影响程度，同时具有较高的准确性。 |
| 关键词: 合成孔径雷达 射频干扰 干扰影响程度评估 区域特征提取 强度特征提取 卷积神经网络 |
| DOI：DOI:10.3969/j.issn.1672-2337.2024.04.005 |
| 分类号:TN974 |
| 基金项目:国家自然科学基金（No.62101096，62171084） |
|
| SAR Radio Frequency Interference⁃Region Intensity Feature Extraction and Joint Evaluation Network |
|
ZHANG Chi, AN Hongyang, LOU Mingyue, LI Zhongyu, WU Junjie, YANG Jianyu
|
|
University of Electronic Science and Technology of China, Chengdu 611731, China
|
| Abstract: |
| Due to the widespread presence of radio frequency signals, synthetic aperture radar （SAR） is susceptible to various radio frequency interference （RFI） during the imaging process, which can lead to a decrease in the quality of SAR images obtained and have a significant impact on subsequent information extraction and target recognition processes. Therefore, it is particularly important to measure the degree of radio frequency interference in SAR images. However, the robustness of existing evaluation methods is usually low, and the size of the region affected by RFI on SAR images is not considered during evaluation. Therefore, a SAR radio frequency interference region?intensity feature extraction and joint evaluation network is proposed in this article. The proposed network consists of two modules. The interference intensity feature extraction module is used to extract the interference intensity information from the input SAR image, while the interference region feature extraction module focuses on the obtaining interference area size and boundary information. Due to the generally large size of SAR images, a multi?level residual and multi?layer feature fusion structure is adopted in the interference intensity feature extraction module to enhance the model’s feature extraction and reuse capabilities. At the same time, in the interference region feature extraction module, emphasis is placed on preserving the most critical region boundary features. In addition, an image dataset of SAR affected by RFI is established to evaluate the effectiveness of the proposed network. The results of comparative experiments show that the network evaluation proposed in this article outperforms other existing methods, and can measure the degree of influence of RFI on SAR images with high accuracy. |
| Key words: synthetic aperture radar（SAR） radio frequency interference（RFI） evaluation of the degree of interference region feature extraction intensity feature extraction convolutional neural network |
