Abstract: The accurate recognition of lamination types is an important application field of optical thin section technology in the process of oilfield exploration and development. In the process of shale reservoir stimulation, due to the unique thin bedding structure and heterogeneity of shale, it is of great significance to accurately identify the lamination types in the formation for selecting the reservoir stimulation location and optimizing the stimulation plan. Compared with the well logging data, the large optical thin sections can achieve more accurate lithology division, and they have more obvious longitudinally continuous lithology variation characteristics than the ordinary thin sections, which can provide reservoir information at the centimeter level, so as to accurately classify the lamination types and optimize the engineering sweet spots. A lamination classification model (referred to as the CNN model) was constructed based on a convolutional neural network (CNN), and three types of lamination were classified and recognized by longitudinally continuous large optical thin section data. The results show that the CNN model can accurately identify fine sand lamination, silty sand lamination, and argillaceous lamination, and the classification accuracy can reach 73% and it is better than that of the YOLOv5 model. The results show that the CNN model can effectively realize intelligent lamination recognition and can deal with complex background and fine lamination features, which provides an efficient and accurate solution for the fine characterization and development of shale oil and gas.