<p>This paper proposes a multicomponent signal recognition and parameter measurement approach based on the Spatial Feature Enhancement Module-YOLOX (SFEM-YOLOX) network, allowing the recognition and parameter measurement of intentionally modulated signals in radar pulses. This method uses signal time-frequency images as input and combines the Convolutional Block Attention Module (CBAM) with basic convolutional modules to enhance the focus of the network on key features. Moreover, it incorporates the Spatial Feature Enhancement Module (SFEM) to capture inter-channel dependencies and specific positional information of feature maps. It also adopts the SFEM-YOLOX object detection network to simultaneously perform signal recognition and parameter measurement. Experiments are then conducted. The obtained results demonstrate that, when the Signal-to-Noise Ratio (SNR) is greater than or equal to -6 dB, the recognition rate reaches more than 90% for all the signal combinations involving three components. For the measurement of parameters of multicomponent signals, the Normalized Mean Square Error (NMSE) is below <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4182_Article_IEq1.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="82" /> </InlineMediaObject> <EquationSource Format="TEX">$7.72 \times 10^{-4}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>7.72</mn> <mo>×</mo> <msup> <mn>10</mn> <mrow> <mo>-</mo> <mn>4</mn> </mrow> </msup> </mrow> </math></EquationSource> </InlineEquation> for SNR greater than or equal to -2 dB. For single-component signals, the NMSE remains below <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4182_Article_IEq2.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="82" /> </InlineMediaObject> <EquationSource Format="TEX">$1.20 \times 10^{-4}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>1.20</mn> <mo>×</mo> <msup> <mn>10</mn> <mrow> <mo>-</mo> <mn>4</mn> </mrow> </msup> </mrow> </math></EquationSource> </InlineEquation> when the SNR is -6 dB. Furthermore, in most of the considered cases, the proposed method significantly outperforms existing approaches.</p>

Multicomponent radar signal recognition and parameter measurement based on SFEM-YOLOX network

期刊论文
发表日期：

2025年5月28日

查看全文

Yihan Xiao,
Yan Wang,
ChangHai Qin,
Tao Chen

摘要

This paper proposes a multicomponent signal recognition and parameter measurement approach based on the Spatial Feature Enhancement Module-YOLOX (SFEM-YOLOX) network, allowing the recognition and parameter measurement of intentionally modulated signals in radar pulses. This method uses signal time-frequency images as input and combines the Convolutional Block Attention Module (CBAM) with basic convolutional modules to enhance the focus of the network on key features. Moreover, it incorporates the Spatial Feature Enhancement Module (SFEM) to capture inter-channel dependencies and specific positional information of feature maps. It also adopts the SFEM-YOLOX object detection network to simultaneously perform signal recognition and parameter measurement. Experiments are then conducted. The obtained results demonstrate that, when the Signal-to-Noise Ratio (SNR) is greater than or equal to -6 dB, the recognition rate reaches more than 90% for all the signal combinations involving three components. For the measurement of parameters of multicomponent signals, the Normalized Mean Square Error (NMSE) is below $7.72 \times 10^{-4}$ $7.72 \times 10^{- 4}$ for SNR greater than or equal to -2 dB. For single-component signals, the NMSE remains below $1.20 \times 10^{-4}$ $1.20 \times 10^{- 4}$ when the SNR is -6 dB. Furthermore, in most of the considered cases, the proposed method significantly outperforms existing approaches.