Image analysis of carbonate clastic rock thin sections using AI systems

The paper deals with the application of machine learning and computer vision technologies for solving the problem of estimating the content of clastic component in carbonates based on thin sections. The training collection is represented by 122 monochrome micro-images of thin sections (fragments of 0.6 x 0.6 mm size) of slightly altered carbonate rocks, divided into two classes - without lithoclasts (lithoclasts are absent or occupy less than 10 % of the image area), with lithoclasts (lithoclasts occupy more than 30 % of the image area). When training the model for image classification, an accuracy of more than 90 % is achieved. The application of the model to the images of thin sections is implemented through console programmes using the Core ML framework. The programmes allow estimating the variations of the “distribution density” of lithoclasts along the profile through the thin section image and to construct a “map” of the distribution of areas with lithoclasts in the image. The resulting data can be used for comparison with geochemical and other numerically expressed information, as well as for selection of areas with the lowest content of allochthonous component in the thin section for the geochemical studies. The model in Core ML format is available upon request from the authors.