Multiscale Fusion of Heterogeneous SAR and Radiometer Data for Robust Maritime Target Enhancement

Abstract

To overcome the limitations of single-modality imaging in maritime ship detection, this article proposes a multiscale active–passive fusion framework. We develop a Laplacian-pyramid-based model to synergistically exploit complementary scattering and radiation information from heterogeneous active and passive data. The framework incorporates a soft ROI mask and a sea-clutter suppression mechanism to enhance target saliency and suppress background interference. A multidimensional evaluation framework with Pareto-frontier analysis is employed to determine the optimal operating point. Experimental results demonstrate that the proposed method not only outperforms single-modality imaging and representative methods including discrete wavelet transform (DWT), Laplacian pyramid (LP), convolutional neural network (CNN), and Transformer in target enhancement, boundary fidelity, and background suppression, but it also shows clear advantages when compared with PCA-FLF from the literature on SAR and microwave data fusion. Furthermore, we apply the proposed fusion method to a downstream binary ship classification and detection task based on the YOLO26 network, further validating its effectiveness in real detection scenarios. Ablation studies also confirm the contribution of each module within the proposed framework. The proposed active–passive fusion framework and analytical methodology provide a robust theoretical foundation for maritime ship detection under complex environments and adverse sea conditions.