Clustering Method for Line Spectra of Ship-Radiated Noise Using Radial Velocity Estimation

A clustering method for line spectra based on radial velocity estimation is proposed to address the challenge of classifying line spectra from multiple ship targets. The method comprises four key stages. Initially, segmented spectral estimation and line spectrum detection are conducted on data received from a single hydrophone to identify the frequencies of interest. Subsequently, the cross-spectrum between each segment and a reference segment is computed, and the real part of the cross-spectrum corresponding to each detected line spectrum frequency is extracted to form a time series. This time series is performed by a time-frequency analysis method. Following this, the frequency with the highest energy at each time interval is identified from the time-frequency analysis result, referred to herein as the time-frequency feature. These features are further refined using a least square fitting algorithm to eliminate outliers. Finally, the radial velocity of each line spectrum is calculated using the optimized time-frequency feature. The Euclidean distances between the radial velocity estimates of different frequency line spectra are utilized for multi-target line spectrum clustering. Experimental results demonstrate that the proposed method effectively clusters line spectra from multiple ship targets, with the clustering outcomes directly applicable to line spectrum classification, multi-target track association, and other relevant scenarios. This approach provides a robust technical solution for multi-target passive detection in complex marine environments.