A Novel ICEEMDAN-Based Denoising Approach Coupled with Multi-Scale Entropy Features for Ship-Radiated Noise Recognition

Effective ship-radiated noise(SR-N) classification is crucial for underwater target detection. This paper introduces a novel approach by applying fine-sorted dispersion entropy (FDE) to SR-N recognition for the first time and extending it to multi-scale features(MFDE) to capture complex temporal characteristics. But SR-N are often contaminated with significant noise, presenting challenges for accurate classification. To address this, we propose several denoising strategies based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN). These strategies are evaluated using MFDE, and the best-performing strategy is selected. This method is then integrated with additional multi-scale entropy features, such as sample entropy. Experimental results show that retaining the first eight IMFs provides the best performance compared to other strategies, such as selecting the IMF with the highest energy or removing the lowest energy IMF. Additionally, MFDE achieves optimal classification performance at scale factor 1, while MSAE demonstrates superior robustness at scale factors 2–3, and MPE dominates when scale factors exceed 3. Overall, the proposed framework, which combines the optimal ICEEMDAN-based denoising strategy with multi-scale entropy analysis, offers an effective solution for improving SR-N classification performance.