Scholarlink

Low Probability of Intercept (LPI) radar avoids non-cooperative receiver from detecting and intercepting its signals by special waveform transmission. It provides numerous waveforms of modulation with variable parameter to enhance anti-interception performance. However, detecting LPI radar signal decreases Signal to Noise Ratio (SNR) due to signal power is less than noise power. In this research, the Parametric Exponential Linear Unit Convolutional Neural Network (PELU-CNN) is proposed for LPI radar waveform. Initially, 12 kinds of LPI radar signals are used to evaluate PELU-CNN performance. An improved Fourier-based Synchro Squeezing Transform (FSST) is used which converts radar signal into Time-Frequency Images (TFI) that represent better performance in low SNRs. Then, a ResNet50 is applied to extract deep features from every resolution channel. At last, PELU-CNN is performed to enhance the fault tolerance of classification. The proposed PELU-CNN achieves a better accuracy of 98.56% for − 6 dB compared to existing method like multi-resolution deep feature fusion method.

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

Detection and Classification of Low Probability of Intercept Radar Signals Using Parametric Exponential Linear Unit Convolutional Neural Network

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Gollara Siddappa Nijaguna,
  • Rohita Yamaganti,
  • Rekha Phadke,
  • N. V. J. Devi Kosuru,
  • R. Archana Reddy

摘要

Low Probability of Intercept (LPI) radar avoids non-cooperative receiver from detecting and intercepting its signals by special waveform transmission. It provides numerous waveforms of modulation with variable parameter to enhance anti-interception performance. However, detecting LPI radar signal decreases Signal to Noise Ratio (SNR) due to signal power is less than noise power. In this research, the Parametric Exponential Linear Unit Convolutional Neural Network (PELU-CNN) is proposed for LPI radar waveform. Initially, 12 kinds of LPI radar signals are used to evaluate PELU-CNN performance. An improved Fourier-based Synchro Squeezing Transform (FSST) is used which converts radar signal into Time-Frequency Images (TFI) that represent better performance in low SNRs. Then, a ResNet50 is applied to extract deep features from every resolution channel. At last, PELU-CNN is performed to enhance the fault tolerance of classification. The proposed PELU-CNN achieves a better accuracy of 98.56% for − 6 dB compared to existing method like multi-resolution deep feature fusion method.