Few-shot point cloud semantic segmentation plays a fundamental role in the computer vision community since annotating point cloud data is quite time-consuming and labor-intensive. Current semantic segmentation methods employ few-shot learning to reduce dependence on labeled samples and enhance model generalization to new categories. Due to the complex 3D geometries of point clouds, significant feature variations exist even within the same category, meaning that a few training samples (support set) might not fully capture all category features. This discrepancy leads to differences in distribution between the support set and the samples used to evaluate the model (query set), impacting the effectiveness of traditional semantic segmentation approaches. In our paper, we employ a prototype enhancement strategy for few-shot point cloud semantic segmentation. Specifically, to align the prototype representation from the support set more closely with the query set, our framework proposes two modules to enhance the generated original prototype, we have developed a Cross Feature Enhancement module, which enhances support set features by reducing differences in terms of distribution of support and query sets. Moreover, we proposed a prototype correction module to refine the prototypes with the aim of matching query sets accurately. We conducted thorough experiments demonstrates the state-of-the-art performance of our model on publicly available benchmarks including S3DIS and ScanNet.