<p>Ship re-identification in complex marine environments faces significant constraints due to the scarcity of training data. The appearance of the same ship varies significantly under different views, lighting conditions, and backgrounds. Existing methods primarily focus on feature enhancement to mitigate sample scarcity. However, most models rely on single-feature spaces, making it difficult to explicitly distinguish identity-related information from interference introduced by imaging conditions. Consequently, intra-class feature dispersion and insufficient retrieval stability often occur even under cross-view matching and complex backgrounds. To address this problem, this paper proposes a foreground guided identity and view disentanglement network, named FIVD-Net, for ship re-identification with few samples. This method first uses a foreground guided branch to generate guidance for the hull region, thereby suppressing the interference of background noise on feature learning. Subsequently, identity and view branches are constructed to model stable identity information and variable view-dependent features separately, while decoupling constraints are applied to reduce semantic overlap between them. Furthermore, a cross-view prototype alignment strategy is introduced to aggregate representations of the same identity across different views near a shared prototype, thereby enhancing intra-class compactness and inter-class separability under few-shot conditions. Experimental results show that FIVD-Net achieves superior performance than the baselines on both public datasets, and demonstrates greater robustness in cross-view retrieval and complex background scenarios.</p>