Abstract: Intelligent lithology prediction is of great importance in geological drilling, capable of improving exploration and mining efficiency, as well as the quality of results. In this study, a method of intelligent lithology prediction while drilling was proposed based on the vibration signals produced by the drill bit breaking rocks during drilling. Specifically, seven types of rocks with the same size and different lithologies were selected, and a micro-drilling experiment was designed to apply different drilling rates and rotary speeds to the rocks, in order to collect the triaxial vibration signals while drilling under multiple drilling conditions. The signals were preprocessed to filter out the interference information and generate the time-frequency images that characterize the signal’s time-frequency domain features through short-time Fourier transform. Then, multiple data augmentation techniques were used to increase the number of images and establish a database to enhance the robustness and generalization ability of the model. The VGG11 (Visual Geometry Group) convolutional neural network algorithm in deep learning was modified, and the database was divided into the training set and test set at a proportion of 8∶2. The effective image information of the training set was extracted, learned and iteratively trained, to obtain an intelligent lithology prediction model. Meanwhile, the three hyperparameters of the model (learning rate, batch size, and iteration times) were continuously adjusted to fit the loss function curve of the training set and the test set and thereby improve the model's prediction accuracy. Finally, the model was evaluated with multiple indicators on the test set. The experimental results showed that: the intelligent lithology prediction model trained based on the vibration data while drilling has strong generalization ability and high prediction accuracy, with an ultimate overall lithology prediction accuracy of 96.85%. Besides, the impact of the dataset size on lithology prediction accuracy was also discussed herein. Moreover, different drilling conditions could cause certain regular changes in the vibration signals while drilling, the rock properties could also cause changes in the vibration signals in the tri-axis direction, and the X, Y and Z axis signals could characterize different processes of the drill bit breaking rocks during drilling. Generally, the intelligent real-time lithology prediction method proposed in this study provides a basis and reference for lithology prediction in practical drilling engineering.