Abstract: Coastal multi-source sensors typically capture ship information independently and often operate under heter-ogeneous frames of discernment, making it difficult to directly utilize source-domain knowledge for accurate ship recognition in the target domain. To address this issue, we propose a collaborative ship recognition meth-od tailored for heterogeneous frames of discernment. First, a cross-domain sample association model is con-structed to link target-domain samples with source-domain samples via few-shot contrastive learning, laying the foundation for the subsequent correction of target-domain confidence scores. Second, a cross-domain con-fidence transfer matrix is designed to unify the frames of discernment across domains, thereby facilitating the auxiliary correction of confidence scores for the associated samples. Finally, a source-knowledge-assisted col-laborative recognition strategy is developed. By dynamically fusing the confidence scores of the associated source-domain samples with those of the target-domain samples, this strategy effectively transfers source-domain knowledge to the target domain. Experiments conducted on a multi-modal ship image dataset demon-strate that the proposed method can successfully leverage source-domain knowledge from heterogeneous frames of discernment to achieve highly accurate recognition in the target domain.