Seminar

Rock classification provides essential information in various engineering geology applications. Researchers have been striving to train AI models based on computer vision (CV) technology to instantly and precisely perceive rock images just as human beings observe rocks. This presentation will briefly introduce (1) the establishment of a large rock image database (more than 16,000 images to be input to the AI models) of five major Hong Kong rock types (fine-grained granite, medium-grained granite, coarse-grained granite, fine ash tuff, and coarse ash tuff), (2) the training, validation, and testing procedures and results (with the precision, recall, and f1-score all above 85%) of different convolutional neural network (CNN) models, (3) supplementary studies based on a suite of CV techniques for distinguishing rock images with different textures, (4) limitations and future prospect of rock image classification.