基于改进SpectralNet的云南40米射电望远镜RFI聚类研究  

RFI clustering study of Yunnan 40 meter radio telescope based on improved SpectralNet

作　　者：贺方彤 梁波 HE Fangtong;LIANG Bo(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650504,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650504

出　　处：《陕西理工大学学报（自然科学版）》2024年第6期36-45,86,共11页Journal of Shaanxi University of Technology:Natural Science Edition

基　　金：国家自然科学基金项目(12063003)。

摘　　要：为了在进行脉冲星观测受到严重射频干扰(RFI)污染问题时制定特定的缓解策略,提出了一种基于卷积自编码器和谱聚类的图像聚类模型CAE-SpectralNet。该模型自编码器部分自动从图像中提取关键特征,避免了手动操作的困难,而谱聚类则将特征空间中距离最近的点聚成簇,有助于揭示数据的内在结构信息。通过对云南天文台40米射电望远镜采集的2 000张脉冲星时域和频域图像数据进行实验,结果表明,SpectralNet模型在改进自编码器结构之后,对比原始SpectralNet模型以及一些传统聚类算法在聚类内部指标上有明显提升,并且聚类结果初步实现了对RFI分类的目标。To develop specific mitigation strategies for addressing the issue of severe radio frequency interference(RFI)contamination during pulsar observations,an image clustering model called CAE-SpectralNet,based on autoencoders and spectral clustering,has been proposed.The autoencoder part of this model automatically extracts key features from images,avoiding the difficulties of manual operations.Spectral clustering clusters the closest points in feature space,which helps to reveal the underlying structural information of the data.Experiments were performed on 2000 pulsar time domain and frequency domain image data collected by the Yunnan Observatory's 40-meter radio telescope.The results show that after improving the structure of the autoencoder,the SpectralNet model significantly improves on the internal indicators of clustering compared to the original SpectralNet model and some traditional clustering algorithms.The clustering results preliminarily achieve the goal of RFI classification.

关 键 词：CAE-SpectralNet模型 图像聚类 RFI的形态归类 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]