Mechanical and electrical construction involves a multitude of components with extended lengths and intricate spatial arrangements, potentially exacerbating construction challenges and the likelihood of installation deviations. To optimize construction management efficiency, we proposed a framework for automated mechanical and electrical construction inspection leveraging panoramic simultaneous localization and mapping (SLAM) and instance segmentation. First, we extracted the coordinates of all panoramic images along the inspection path using the OpenVSLAM algorithm. Second, the PointRend algorithm was employed to segment the mechanical and electrical construction components. Finally, the installation discrepancies were automatically identified by comparing segmentation images of real scenes (as-built) with virtual images in Unreal Engine (as-planned). This framework was successfully deployed in a mechanical and electrical construction project in Shanghai, China, demonstrating precise image segmentation and accurate identification of installation discrepancies. The study establishes a research paradigm for vision-based intelligent construction inspection, which can be further expanded to encompass large-scale indoor construction progress tracking and quality monitoring.