Multiscale cumulative residual dispersion entropy: a robust nonlinear feature extraction framework for underwater target recognition

Underwater target recognition faces great challenges due to the complex acoustic interference and non-stationary characteristics of hydroacoustic signals. To address this problem, we propose a novel multiscale cumulative residual dispersion entropy (MCRDE) framework that integrates multiscale analysis, cumulative residual entropy theory, and dispersion model to quantify the nonlinear dynamics of underwater targets. Specifically, MCRDE overcomes the limitations of traditional entropy-based methods by (1) jointly characterizing signal complexity at multiple time scales, (2) enhancing robustness to noise through cumulative residual operators, and (3) capturing hierarchical dynamic features through dispersive mode mapping. Experimental results on real hydroacoustic datasets show that the classification accuracy of MCRDE is improved by 15.2% compared to the refined composite multiscale dispersion entropy (RCMDE), and the classification accuracy of MCRDE is improved by 15.2% compared to the traditional multiscale fluctuation-based dispersion entropy (MFDE). The proposed framework provides a generalizable tool for underwater target characterization in defense and ocean exploration applications.