Scholarlink

<p>Wideband high frequency (HF) signal detection is crucial for radio intelligence operations. However, multiple signals coexisting within a wideband spectrum, combined with dispersive and time-varying HF channels, complicates signal detection. This paper proposes a novel method called High Frequency-You Only Look Once (HF-YOLO) to address the above problems. Specifically, HF-YOLO processes wideband time-frequency image (WTFI) as input. An Inception feature extraction (IFE) module, based on the Inception architecture and group convolution, is proposed to improve the modeling of strip-distributed signals in WTFI while reducing computational complexity. Additionally, a time-frequency attention (TFA) module is introduced and improved considering the narrow bandwidth and strong temporal correlation of HF signals. The TFA effectively suppresses out-of-band interference and improves the detection of weak signals. To evaluate performance, a dataset called wideband time-frequency image dataset (WTFID) is constructed comprising a variety of HF signals covering diverse communication scenarios. Experimental results demonstrate that HF-YOLO achieves high detection accuracy and real-time performance even in signal-to-noise ratio (SNR) below 0 dB, confirming the effectiveness of HF-YOLO under complex channel conditions.</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

HF-YOLO: A Model for High Frequency Signal Detection and Recognition in Wideband

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Libin Gu,
  • Tianxue Chen,
  • Kai Liu

摘要

Wideband high frequency (HF) signal detection is crucial for radio intelligence operations. However, multiple signals coexisting within a wideband spectrum, combined with dispersive and time-varying HF channels, complicates signal detection. This paper proposes a novel method called High Frequency-You Only Look Once (HF-YOLO) to address the above problems. Specifically, HF-YOLO processes wideband time-frequency image (WTFI) as input. An Inception feature extraction (IFE) module, based on the Inception architecture and group convolution, is proposed to improve the modeling of strip-distributed signals in WTFI while reducing computational complexity. Additionally, a time-frequency attention (TFA) module is introduced and improved considering the narrow bandwidth and strong temporal correlation of HF signals. The TFA effectively suppresses out-of-band interference and improves the detection of weak signals. To evaluate performance, a dataset called wideband time-frequency image dataset (WTFID) is constructed comprising a variety of HF signals covering diverse communication scenarios. Experimental results demonstrate that HF-YOLO achieves high detection accuracy and real-time performance even in signal-to-noise ratio (SNR) below 0 dB, confirming the effectiveness of HF-YOLO under complex channel conditions.