Spatial domain identification is a fundamental task in spatial transcriptomics, aiming to partition tissues into coherent regions that reflect both transcriptional similarity and spatial organization. Existing methods typically generate low-dimensional embeddings through predefined transformations or neural network encoders, and treat these representations as fixed inputs for downstream clustering. This limits the model’s ability to adapt to multiple spatial and expression-derived constraints. To address this limitation, we propose Graph-Regularized Embedding Refinement for Spatial Transcriptomics (GRER-ST), a refinement framework that enhances spatial domain identification by explicitly optimizing the latent embedding. GRER-ST first integrates spatial coordinates and gene expression profiles to construct an initial low-dimensional representation that captures local spatial context. A trainable latent embedding, initialized from this representation, is then refined under a unified objective that incorporates spatial smoothness, expression similarity, and neighborhood-based contrastive regularization. By refining the embedding rather than relying on a fixed representation, GRER-ST produces spatial domains that are both locally coherent and discriminative. Extensive experiments on multiple spatial transcriptomics datasets demonstrate that GRER-ST consistently outperforms baseline methods in spatial domain identification, yielding more spatially coherent domain structures. These findings suggest that embedding refinement is an effective and flexible strategy for integrating heterogeneous spatial and transcriptional information in spatial transcriptomics.