<p>Lithology identification is an important basis for guiding tunnel engineering construction as well as estimating the reserves of oil, gas, and solid mineral resources. Drilling technology is one of the key methods for quickly acquiring lithological information. This paper proposes a method for identifying lithology in drilling cuttings by combining object detection algorithms with visible and near-infrared (VNIR) spectroscopy. Firstly, a laboratory VNIR spectral dataset containing 27 lithologies was established, and a classification model based on the YOLOv5s algorithm was developed specifically for identifying spectral data features in the laboratory setting. Then the field spectral data for five lithologies were collected, and the transfer learning method was employed to achieve intelligent lithology identification in field drilling cuttings. Finally, this method was applied to a real-world engineering project. The results show that the identification accuracy for laboratory lithology data is nearly 99%, with evaluation metrics reflecting the model’s excellent performance. The accuracy of the transfer learning method used for the intelligent identification of field drilling cuttings data reached 98%. In the engineering application, only one of the 30 cuttings samples was incorrectly identified, and the identification results were generally consistent with the excavation verification outcomes. The YOLOv5s object detection algorithm demonstrates good identification performance for spectral curves of different lithologies. This method holds promise for wide application in geological exploration, oil and gas resource development, and mining industries.</p>