Integration of ICEEMDAN-Wavelet Thresholding Joint Denoising with GWO-LSTM for Detection

This study addresses the critical issue of weak target recognition in intricate maritime environments, hindered by significant sea clutter, by proposing a fusion detection approach that utilizes ICEEMDAN-wavelet threshold signal reconstruction and GWO-LSTM. Initially, the ICEEMDAN algorithm, incorporating a dual noise injection mechanism and a dynamic stopping criterion, is employed to effectively mitigate modal aliasing via adaptive noise compensation, with experimental results indicating a substantial reduction in the modal orthogonality index (IOO) compared to traditional decomposition algorithms. Subsequently, joint wavelet threshold denoising is implemented to establish a dynamic equilibrium between noise suppression and feature preservation through the collaborative effects of modal decomposition reconstruction and wavelet threshold shrinkage. Finally, the Grey Wolf (GWO) optimization strategy is applied to the kernel KELM classifier for the identification of target presence within the signal. The novel technique significantly mitigates marine clutter while preserving its chaotic characteristics. Under identical conditions, the training duration is reduced, the prediction accuracy is elevated, and the algorithm demonstrates significant advantages in identifying small targets amidst sea clutter compared to the conventional chaotic prediction method.