Near-infrared multispectral imaging (NIR-MSI) has attracted attention in the surgical field owing to its ability to penetrate tissues and analyze molecular vibrations with high spatial resolution, thus enabling the identification of deep or similarly colored tissues. However, conventional devices have limitations such as non-portability and long imaging times, rendering the surgical use of NIR-MSI challenging. This study developed a prototype NIR-MSI rigid endoscope system that can perform high-speed imaging and tissue classification. This system was constructed using a custom-made rigid endoscope that can relay NIR images, an InGaAs camera, and a lighting device containing nine high-power LEDs within the range of 940-1,535 nm. The developed device enables continuous processing of light irradiation, data acquisition, neural network analysis, and result display using customized software. The performance of the system was verified by testing tissue transparency and component classification. Tissue transparency test showed that text written on a letterboard could be recognized through a 5-mm thick layer of chicken breast meat. The system completed the imaging and neural network analysis of four color-matched transparent resins, which could not be distinguished readily under visible light, within 2 s, with an average accuracy of 96%. The proposed system may be applied for identifying deep or similarly colored tissues in the setting of non-invasive and real-time endoscopic surgery.