Abstract: The objective of this study was to improve ship target recognition performance through a multimodal fusion approach， leveraging the strengths of different sensors. Traditional unimodal recognition methods typically rely on a single sensor’s features and achieve high accuracy under specific conditions. However， these methods are limited in their ability to capture multidimensional information about the target and exhibit variability in feature performance across different recognition scenarios. While multimodal fusion approaches can address these shortcomings， most existing methods primarily use feature concatenation or weighted fusion， which fail to fully exploit the complementarity and correlation between features from different modalities. Additionally， they do not adequately consider the impact of target distance on feature extraction， which limits their fusion effectiveness. To address these issues， this study proposed a multimodal fusion ship recognition method based on distance-controlled contrastive learning to enhance the performance of multimodal fusion in ship target recognition. The proposed method involves three novel modules to fully utilize the feature information from both sensors， thereby enhancing the fusion process. First， a distance control module was designed to guide the network in the usage of feature vectors under varying distance conditions， thus optimizing the feature fusion process for different distances. Second， a decoupled contrastive loss module was introduced to preserve the unique information of each modality， thus improving the complementarity between features. Finally， a supervised contrastive loss module was implemented to capture common information from the fused features of targets belonging to the same category， thus establishing the correlation between modality-specific features. The method was validated through experiments using real-world data， with ship targets tested at different distances. The experimental results demonstrated that the proposed distance-controlled radar-infrared feature fusion network outperformed unimodal recognition methods， improving ship recognition accuracy by more than 9% across various distances. Furthermore， it achieved a 4.65% improvement in recognition accuracy compared with existing fusion-based recognition methods. The proposed approach not only enhanced ship target recognition accuracy， but also adapted flexibly to feature variations at different distances， thus demonstrating its superiority in multimodal fusion ship target classification. In conclusion， the proposed method effectively addresses key challenges in multimodal fusion by incorporating distance control， ensuring optimal feature fusion across varying conditions. The use of decoupled contrastive loss helps preserve modality-specific features， while benefiting from the complementary information provided by the other modality. Additionally， the supervised contrastive loss module strengthens the correlation between similar features across modalities， thus improving the classification accuracy. This combination of modules results in a robust and adaptable fusion network， offering significant improvements over existing methods and making it highly effective for multimodal fusion ship target classification.