Scholarlink

<p>The development of the electronic technology led to various types of radar interference. The accurate interference identification is a prerequisite for electronic countermeasures. Conventional deep learning-based radar interference recognition requires a large amount of labeled data training, which cannot be easily obtained in the actual environment. This paper solves the problem by proposing a radar interference recognition algorithm based on semi-supervised learning(SSL), referred to as SSLRIR. Short Time Fourier Transform (STFT) is first performed on the received interference waveform to convert it into a time-frequency image holding distinct features. A semi-supervised recognition framework based on Mean Teacher (MT), incorporating output consistency constraints relying on random erasure data perturbation, is then established. An improved MobileNetV4 is used as feature extraction network, and the Convolutional Block Attention Module (CBAM) is incorporated into the network in order to increase its ability to extract deep features of data at low signal-to-noise ratio. A Redundancy Reduction Block (RRB) is introduced to reduce the redundancy, which allows the network to better understand key features and performs radar interference recognition when dealing with a small amount of labeled data. Afterwards, experiments are conducted, demonstrating that, with only 10% labeled data and a Jammer-to-Noise Ratio (JNR) of -8 dB, the average recognition accuracy for 16 types of composite radar interference reaches 93.19%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Semi-supervised radar interference recognition based on deep learning

  • 期刊论文

  • 发表日期：

    2025年12月13日

查看全文
  • Yihan Xiao,
  • Chao Li,
  • Mengmeng Huang,
  • Qianrong Lu

摘要

The development of the electronic technology led to various types of radar interference. The accurate interference identification is a prerequisite for electronic countermeasures. Conventional deep learning-based radar interference recognition requires a large amount of labeled data training, which cannot be easily obtained in the actual environment. This paper solves the problem by proposing a radar interference recognition algorithm based on semi-supervised learning(SSL), referred to as SSLRIR. Short Time Fourier Transform (STFT) is first performed on the received interference waveform to convert it into a time-frequency image holding distinct features. A semi-supervised recognition framework based on Mean Teacher (MT), incorporating output consistency constraints relying on random erasure data perturbation, is then established. An improved MobileNetV4 is used as feature extraction network, and the Convolutional Block Attention Module (CBAM) is incorporated into the network in order to increase its ability to extract deep features of data at low signal-to-noise ratio. A Redundancy Reduction Block (RRB) is introduced to reduce the redundancy, which allows the network to better understand key features and performs radar interference recognition when dealing with a small amount of labeled data. Afterwards, experiments are conducted, demonstrating that, with only 10% labeled data and a Jammer-to-Noise Ratio (JNR) of -8 dB, the average recognition accuracy for 16 types of composite radar interference reaches 93.19%.