Graph structure learning has long been considered an effective method for optimizing graph data in real world. Unsupervised learning methods are highly regarded because they do not rely on labeled data and specific downstream tasks. Contrastive learning is one of the widely used methods. We combine the two by using a feedback mechanism. To maintain model robustness, adversarial views are added. Considering that adversarial attacks may affect the homophily of graph data itself, we perform homophily-enhanced processing on graph data during graph structure learning to maintain model stability and learn a better graph structure. In this paper, we propose graph structure learning with homophily enhancement under adversarial contrastive learning framework. Specifically, we add the optimized graph structure as a view to the adversarial graph contrastive learning process and use the graph representation trained by contrastive learning to further optimize the graph structure. Extensive experiments on various real-world datasets can prove the effectiveness of our method.