Embedding Soft Thresholding Function into Deep Learning Models for Noisy Radar Emitter Signal Recognition

• DOI: 10.3390/electronics11142142
• Publication Date: 2022-07-08T15:37:08Z
• Authors: Jifei Pan; Shengli Zhang; Lingsi Xia; Long Tan; Linqing Guo
• Keywords: Sigmoid function; SIGNAL (programming language); Feature (linguistics)

Radar emitter signal recognition under noisy background is one of the focus areas in research on radar processing. In this study, soft thresholding function embedded into deep learning network models as a novel nonlinear activation function, achieving advanced results. Specifically, an sub-network used to learn threshold according input feature, which results each feature having its own independent function. Compared with conventional functions, characterized by flexible conversion and ability obtain more discriminative features. By way, noise features can be flexibly filtered while retaining features, thus improving accuracy. Under condition Gaussian Laplacian signal-to-noise ratio −8 dB −2 dB, experimental show that overall average accuracy reached 88.55%, was 11.82%, 8.12%, 2.16%, 1.46% higher than those Sigmoid, PReLU, ReLU, ELU, SELU, respectively.