Rock classification model based on transfer learning and convolutional neural network

Abstract

∗In geological work, some manual discrimination methods used for rock identification tasks are highly influenced by personal subjective factors and have low identification efficiency, and many experiments are performed on pre-processed rock thin sections, which are difficult to meet the requirements of modern geological research work. In order to solve the above problems, this paper proposes a rock classification model based on transfer learning and convolutional neural network using the original rock images as the data set, and finetune the VGG16 convolutional neural network frozen with three convolutional layers using the data expanded rock training set samples using transfer learning method. reached a maximum of 90.24%. The experimental results show that the rock classification model built by finetune and VGG16 in this paper has a short training period and can accurately classify and recognize rocks automatically.