Abstract: To screen out ship noise feature line spectrum, a feature data recognition algorithm for massive data is proposed. The noise spectrum is composed of continuous spectrum and line spectrum, which is treated as the trend term. The power spectrum trend term is obtained by fitting the power spectrum based on the least square principle. Taking the trend term as the quasi-zero baseline, the power spectrum is divided into two parts. The discontinuous spectral lines on the zero line are grouped and the local optimization is carried out to obtain the initial feature line spectrum. The ship noise feature line spectrum is obtained by peak sorting according to the spectral line weight. The algorithm realizes the effective extraction of feature line spectrum. The effectiveness of the algorithm is verified by the measured data, which has a certain practical engineering application value.