The task of community detection aims to identify the latent group structures of nodes in graph data. However, existing methods are often limited in their effectiveness due to an over-reliance on pairwise relationships among nodes while neglecting higher-order structural information. Furthermore, traditional contrastive learning techniques generate augmented views by randomly disrupting graph structures, which can result in loss of original information and diminished node discrimination. To address these issues, this paper proposes a novel method that integrates hypergraph higher-order relationships with non-destructive multi-view contrastive learning. Specifically, we construct a dual-path feature learning framework that dynamically weights node attribute features through a self-attention encoder and employs hypergraph convolution to model the high-order interactions among multiple nodes connected by hyperedges, thus overcoming the limitations of pairwise relationships. We introduce a non-destructive multi-view contrastive strategy that utilizes attribute views and higher-order structural views as contrasting pairs. By maximizing mutual information, we optimize node representations while preserving graph integrity and enhancing representation discrimination, thereby reducing noise interference. Experimental results demonstrate the superior performance of the proposed method across multiple datasets, validating its efficacy and potential applications in community detection tasks.