Radar emitter fingerprint recognition based on bispectrum and SURF feature

Convolutional neural network applied to specific emitter identification based on pulse waveform images

“…7 presents the identification performance of different input features when the SNR ranges from 4 to 36 dB. And these features include our proposed PWI, HHT [7], original instantaneous phase (OIP) [19], NVGE [22], BS [33] and Hilbert spectrum (HS) [34], which are fed to CNN, respectively.…”

“…The bispectrum (BS) is also a tool to present the time‐domain signals. In [33], the BS projection is converted to greyscale, from which to extract the SURF feature and recognise them by feature matching. In [34], a promising SEI approach based on the compressed BS of the received signals using the CNN is proposed, which outperforms other existing schemes.…”

Specific Emitter Identification via Feature Extraction in Hilbert-Huang Transform Domain

“…Various features and combinations are presented to perform signal-level SEI, although the mechanism has not been explained theoretically. High-order cumulants [7], wavelet ridge and high order spectra [8], bispectrum and its variants [9][10][11] have demonstrated the effectiveness for the given conditions. Recently, Hilbert-Huang transform (HHT) has proven the superiority in the unique representation and descriptive ability for SEI [12][13][14][15].…”

Radar Emitter Recognition Based on the Energy Cumulant of Short Time Fourier Transform and Reinforced Deep Belief Network

“…A lot of studies on radar emitter recognition based on the intra-pulse features have been conducted. According to the supplied features, these studies mainly concentrate on the time domain features [ 10 , 12 , 13 ], transformed domain features [ 10 , 11 , 12 , 13 , 14 ] and statistics features [ 15 ]. From the aspect of the recognition method, the studies can be divided into two groups: classifiers based on multi-thresholds and multi-regulations [ 10 , 12 ], and classifiers based on learning algorithms [ 16 , 17 , 18 , 19 , 20 , 21 ].…”

“…Because the foundation of the third category originates from the first one, their recognition and classification methods are similar. In [ 13 , 15 , 21 , 24 ], different features are combined together to increase the uniqueness of representation, which makes the recognition easier compared with before. All the recognition methods aim to attain the high-correct recognition rate and good generalization ability, which depends on whether the chosen features have obvious distinction and anti-noise property, and whether the recognition methods have strong adaptability to the data outside the learning set.…”