An Algorithm Enhancing Line Spectrum in Frequency Domain Based on Stacked Sparse Autoencoder

The noise radiated from marines consists of line spectra and continuous spectra, within which the line spectra contain high energy and propagate over long distances. Traditional methods for line spectrum detection exhibit poor performance in underwater target detection for weak noise signals. Hence, this paper proposes an algorithm enhancing the frequency domain line spectrum based on stacked sparse autoencoders to exploit the sparsity of the frequency and spatial domain line spectrum signals. This algorithm shows stronger line spectrum detection capability than the Conventional Adaptive Line Enhancer algorithm (C-ALE). Ceteris paribus, this algorithm yields an increase of 3.2 dB in the output Signal-to-Noise Ratio (oSNR).