Effective identification of cancer subtypes is crucial for pathological analysis and targeted therapy of cancer. With the rapid advancement of high-throughput sequencing technology, the volume of multi-omics data has increased exponentially. In recent years, multi-omics clustering has garnered significant attention as a method that can effectively integrate multi-omics data and facilitate cancer subtype segmentation. Although existing methods have yielded promising results, they still encounter several challenges: 1) It is difficult to fully explore the consistent associations between cross-omics data and the complementary characteristics within specific omics, and the investigation of the topological structure within specific omics is insufficient; 2) There is a lack of effective strategies for obtaining clustering-friendly consensus matrices. To solve these problems, we propose a multi-omics cancer subtype identification method based on a dual-channel encoder and Cauchy-Schwarz(CS) divergence optimization(DECSD). Specifically, our method processes intra-omics complementary features through a dual-channel encoder, learns cross-omics consistency via a weighted fusion module, and simultaneously captures structural patterns within each omic using Graph Convolutional Network(GCN). Subsequently, we derive high-quality latent representations via a sequential two-stage learning process. Finally, CS divergence is integrated to obtain discriminative consensus representations for clustering. Extensive experimental results on four real-world datasets show that our method surpasses the existing state-of-the-art methods in multi-omics data clustering analysis.