The development of underground spaces such as tunnels, subways, logistics warehouses, and complex facilities is continuing. However, owing to poor planning and reckless expansion, there has also been an increase in underground accidents. As such, it is important to obtain accurate geotechnical data on underground spaces for optimal construction outcomes and to ensure the safety of workers. Borehole cores contain essential geological information towards achieving these ends; however, rock classification using borehole cores takes a long time and the classification depends on the interpreter. To address these issues, we performed rock classification based on borehole sensing images using a convolutional neural network (CNN) combined with deep learning techniques. The data used for the training were collected from images of borehole cores in Hang-dong, Guro-gu, Seoul, and Hyeol-dong, Taebaek, Republic of Korea. We used the collected two datasets: a rod dataset labeled by the rock type of the borehole core rod unit and a grid dataset labeled by the rock type unit. The rock types were classified into basalt, gneiss, limestone, mudstone, and shale. In addition, mixed-rock and loss classes were added to the classifications. For the image classification process, we proposed three methods: general deep-learning-based image classification, multiregion image classification, and multiregion image classification using a scoring process. An experiment was conducted to validate these methods. A maximum accuracy of 99.02% was achieved in the validation process. The proposed methods introduced here are expected to reduce the time and costs associated with creating geotechnical databases.