Building YoloV4 models for identification of rock minerals in thin section

Abstract

Rock mineral identification is a costly and time-consuming task using conventional methods of testing physical and chemical properties, especially in the petrographic laboratory. A comprehensive identification model for three rock minerals in sedimentary rocks based on the YoloV4 model is available as a solution. The models predict rock minerals by calculating the pixels and the weights that have been trained previously. First, the YoloV4 models and framework were built. Then, a total of 44 manually labelled thin section images (sedimentary rocks thin section) were used to create the model to detect minerals accurately. The MAP and loss results showed that the parameters of the minerals detection model in PPL are 11% and 1.19, respectively. Meanwhile, The MAP and loss results of XPL are 19% and 1.18, respectively. Finally, Identification of rock minerals using deep learning algorithms is a very promising idea especially the YoloV4 model can build a comprehensive detection of rock samples in thin sections effectively.