Lithology Identification based on ResNeSt in High-resolution Rock Images

Lithology identification is of great reference value for geological analysis and investigation. This paper presents a method for the classification of high-resolution rock image. For the properties of the rock samples, this paper generates rock stitching images as dataset to train and test the ResNeSt deep learning model. Test results on the relevant data show that 16 rock stitching images are produced with an accuracy rate of 87.1% for hard voting prediction. This method is more competitive in the task of high-resolution rock image classification, which provides a new way to solve the problem of fast lithology classification of rock image in the geological survey.