Core box image recognition and its improvement with a new augmentation technique

Most methods for automated full-bore rock core image analysis (description, colour, properties distribution, etc.) are based on separate column analyses. The is usually imaged in a box because of the significant amount time taken to get an each column. work presents innovative method and algorithm columns extraction from boxes. conditions boxes imaging may differ tremendously. Such differences disastrous machine learning algorithms which need large dataset describing all possible data variations. Still, such images have some standard features - core. Thus, we can emulate different environments with unique augmentation described this work. It called template-like (TLA). tested various environments, results compared trained both 'traditional' mix traditional TLA data. provides better metrics detect most new images, unlike without TLA. implemented description system speeds up processing by factor 20.