MCEF-NET: A Multimodal Contribution Evaluation Fusion Network for Maritime Target Recognition

With the rapid development of low-end maritime devices and shipborne sensors, traditional single-modal recognition can no longer meet the demand for high accuracy in maritime environment. As a result, multimodal learning, which integrates data from different sensors, has gradually become the main approach for maritime target recognition. However, due to variations in the environments and platforms where data is collected, the quantity of useful information provided by each modality differs, and certain modalities may even introduce noise. This discrepancy adversely affects the performance of multimodal fusion recognition. However, most existing multimodal maritime recognition methods overlook these differences, which constrains the performance of the recognition models. To address this concern, we propose a multi-modal contribution evaluation fusion network (MCEF-NET) to achieve efficiency-enhanced fusion for multimodal maritime target recognition. In this model, a Feature Filter Module (FFM) is introduced to effectively suppress irrelevant information, mitigate distribution discrepancies between modalities, and enhance the robustness of multimodal feature extraction. Furthermore, we design a Contribution-Rating Fusion (CRF) mechanism that dynamically allocates fusion weights according to the contribution of each modality’s features, thereby minimizing the influence of low-value modalities on the final fusion performance. The MCEF-NET was evaluated on the publicly available VAIS maritime infrared-visible multimodal dataset, exhibiting superior accuracy and computational efficiency compared to existing state-of-the-art methods.