基于自适应卷积去噪网络的随机噪声压制方法  

Random Noise Suppression Method Based on Convolutional Blind Denoising Network

作　　者：张文征 韩顺利[1] 张亭 朱军锋 ZHANG Wen-zheng;HAN Shun-li;ZHANG Ting;ZHU Jun-feng(The 41st Institute of China Electronic Technology Group Corporation,Qingdao Shandong 266555,China)

机构地区：[1]中国电子科技集团第四十一研究所,山东青岛266555

出　　处：《计算机仿真》2025年第3期294-298,共5页Computer Simulation

基　　金：国家自然科学基金(62127804)。

摘　　要：针对近场微波测试过程中信号的随机噪声,使用深度学习方法,采用自适应卷积去噪网络,实现近场微波信号随机噪声自适应智能压制。利用非对称学习损失函数,构建噪声水平估计算法和卷积去噪子网络,提高数据训练可靠性,增强去噪结果的有效性和准确性。同时采用模型噪声与实际噪声两种类型数据对去噪网络进行交替训练,提升网络模型的去噪效果和泛化能力。与传统随机噪声压制算法相比,上述方法对数据模型和先验信息的依赖性低,可以在未知噪声标准差的情况下实现近场微波数据自适应去噪。A self-adaptive intelligent denoising of near-field microwave signals is achieved by adopting a convolutional blind denoising network(CBDNet)based on deep learning.An asymmetric learning loss function is used to construct noise level estimation and convolutional denoising sub-networks,which enhances the effectiveness and accuracy of the denoising results.Meanwhile,model noise and actual noise data are studied during the continuing data training,in order to improve the denoising effect and generalization ability of the network model.Compared with the traditional random noise suppression algorithm,the self-adaptive intelligent denoising method could realize the adaptive denoising of near-field microwave data under the condition of unknown noise standard deviation with lower dependence on the data model and prior information.

关 键 词：随机噪声 深度学习 自适应卷积去噪网络 噪声估计去噪网络 近场微波信号 

分 类 号：TN911.7[电子电信—通信与信息系统]