<p>Since the emergence of Neural Radiance Fields, neural implicit surface reconstruction methods have achieved remarkable progress. However, the reconstruction process still struggles with precision, often leading to blurry regions in the fine details. To address this challenge, we propose Feature-NeuS, a neural implicit surface reconstruction method that integrates multi-view semantic features consistency. We extract multi-scale features that include high-level semantic representations and low-level visual information, which serve as prior knowledge for the network to refine the fine details in the 3D model. Additionally, we introduce a multi-view consistency loss for surface points, employing multi-view geometric constraints on image level and feature level to enhance the accuracy and sharpness of the reconstruction. Our method aims to better capture and refine the intricate structures of the 3D reconstructed model by utilizing both image-derived features and multi-view consistency. Extensive qualitative and quantitative experiments demonstrate that our method outperforms the state-of-the-art on DTU, BlendedMVS, and Tank &amp; Temple datasets, particularly in recovering fine model details.</p>