Owing to the unique material properties and complex geometry of on-orbit formed space truss structures, traditional 3D visual inspection methods often fail to achieve complete model reconstruction, thereby compromising defect detection accuracy. This paper presents a visionbased detection method for truss structures that leverages 3D curve network graph optimization and designs a rotating visual scanning system for 3D inspection. This system enables comprehensive structural reconstruction and precise defect localization, addressing challenges in on-orbit manufacturing quality control. First, image curve feature recognition is combined with a curve feature matching approach that incorporates distance criteria and curve consistency constraints to establish correspondence between weak-textured slender objects across sequential image frames. Second, by leveraging the principle of 3D curve network graph structure optimization, the method iteratively updates the camera pose and the target structure, thereby computing and refining the geometric topology of the truss while estimating member diameters. Furthermore, a defect detection method based on 3D curvature analysis is introduced and complemented with 2D image validation, enabling accurate identification of typical defects such as virtual joints and breakpoints. Experimental results indicate that, compared to traditional point-feature-matching-based 3D reconstruction methods, the proposed approach achieves superior reconstruction quality and efficiency for slender truss structures. Both geometric parameters and defect detection accuracy exceed 0.3 mm, meeting the requirements for on-orbit manufacturing site inspections. Notably, this method does not rely on external calibration or background features, providing a robust technological foundation for quality control in the on-orbit manufacturing of complex space structures.