This paper identifies a gap in existing methods regarding the comprehensive consideration of consistency constraints in the learning process of cross-modal hash functions. To address this, the paper proposes the construction of domain consistency, semantic consistency, and modal consistency losses based on the previously established structural consistency loss function. These constraints guide the learning process of the cross-modal hash synthesis objective function, thereby enhancing unsupervised cross-modal retrieval optimization. The key innovations lie in the introduction of a domain adaptive learning method that integrates consistency constraints on structural information, domain features, semantic distribution, and modal feature differences, alongside the construction of a multi-metric relationship matrix to leverage diverse data representations and improve feature representation consistency. We conducted experiments in a fair environment and compared the results with six relevant papers proposed in recent years, and the results showed that our method has superiority.