Silencers are key gene regulatory elements that control gene expression and maintain genomic stability. Dysregulation of silencers is linked to diseases like cancer and genetic disorders. Traditional silencer identification relies on manually extracted features, which are time-consuming, costly, and inefficient. Moreover, most studies focus on local DNA sequence features, neglecting contextual relationships that limit model accuracy. To address these issues, we propose DCFICSH, a deep learning framework that employs dual-channel fusion with distinct automated encoding methods for DNA sequences, incorporating omics data as an additional modality. This multi-modal approach enriches the model’s understanding by integrating both DNA sequence and omics data, allowing for a more comprehensive identification of silencers by capturing both local and contextual features. We enhance feature extraction with BiGRU, which captures sequence dependencies, and Transformer, which extracts long-range dependencies via self-attention. DCFICSH improves AUROC and AUPR by 1% to 2% over existing methods for silencer recognition in HepG2 and K562 cell lines. Additionally, we have developed a public web server that offers all relevant resources for this study, enabling researchers to quickly perform silencer predictions. The source code can be accessed at https://github.com/1Yjd/DCFICSH.git .