| 摘要: |
| 识别复杂无线通信环境中的无线频谱状态是频谱监测系统的核心任务之一。面对频谱观测时间限制导致的数据不足和类别不平衡问题，提出了基于原型校正网络的小样本学习框架，重点集中在提取局部特征上。具体地，通过采用多层卷积块提取各个频谱状态的局部描述符，每个描述符代表不同类别的特征。此步骤有助于捕捉频谱状态的关键特征，同时有效应对数据不足的问题。其次，引入了一个基于原型网络的分类器，该分类器使用提取的局部描述符对频谱数据进行状态分类。原型网络的设计能够更好地处理频谱观测数据类别不平衡的问题，确保每个类别都得到适当的关注。实验结果表明，所提出的原型校正网络框架相对于传统方法在无线频谱识别准确度上提升了5%，在信噪比为-8 dB时仍实现了超过75%的准确率。 |
| 关键词: 无线频谱状态识别 频谱状态地图 小样本学习 原型校正网络 局部描述符 |
| DOI：10.20079/j.issn.1001-893x.240319002 |
|
| 基金项目:国家自然科学基金资助项目（62371082） |
|
| Local Descriptor Aided Prototype Correction Network Based Wireless Spectrum Status Recognition |
| LI Yue,SHEN Bin,WANG Xin |
| (1.School of Communications and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;2.Chongqing Key Laboratory of Mobile Communications Technology,Chongqing 400065,China) |
| Abstract: |
| Identifying wireless spectrum status in complex wireless communication environments is one of the core tasks of spectrum monitoring systems.For the problems of insufficient data and imbalanced categories caused by time constraints in spectrum observation,the authors propose a few shot learning framework based on prototype correction networks,focusing on extracting local features.Specifically,by using multi-layer convolution blocks,local descriptors of each spectrum status are extracted,and each descriptor represents a different category of features.This step facilitates capturing key features of the spectrum status while effectively coping with the problem of insufficient data.Secondly,a prototype network based classifier is introduced that uses extracted local descriptors to perform status classification on spectrum data.The prototype network is designed to better handle the problem of imbalanced spectrum observation data categories,ensuring that each category receives appropriate attention.Experimental results show that the proposed prototype correction network framework has achieved a 5%increase in wireless spectrum recognition accuracy compared with traditional methods and an accuracy of over 75% is achieved even at -8 dB signal-to-noise ratio. |
| Key words: wireless spectrum status recognition spectrum status map few shot learning prototype correction network local descriptor |
