Due to its excellent storage and retrieval performance, Cross-Modal Hashing (CMH) has garnered widespread attention in recent years. However, current methods still face two major limitations: 1) most methods fail to adequately address noise in the labels, 2) many methods neglect learning the correlations between labels. To address these issues, we propose Label-enhanced Cross-modal Hashing with Dual-semantic Learning (LCHDL) in this study. LCHDL mainly consists of two parts, i.e., DeNoise Label Semantics Learning (DNLSL) and Static Label Correlations Embedding (SLCE). DNLSL uses low-rank and sparse representation to reduce noise containing in the labels. SLCE employs a relaxation strategy to achieve static label correlations embedding. Experiments on two benchmark datasets demonstrate that the performance of LCHDL surpasses many state-of-the-art methods. The source code for LCHDL is publicly available at https://github.com/Fizzyf/LCHDL .