As the value of oak wood increases, it is necessary to think of solutions that help make use as much material as possible during the industrial processing of wood. In order to study the possibilities of efficient use of wood, the specific oak wood features and defects were compiled from images obtained on a scanner prototype. The scanner, based on two TeleDyne Linea line cameras, captured high-quality images at 17,000 by 900 resolution. When the images were marked, two different mask r-cnn models were trained, using the instance segmentation method. The original model (first) contained only the data obtained from the scanner, while the training of the second model used the original images and additional images that were artificially adjusted to expand the data set. New images where generated using keras framework. There were only two data augmentation methods used like "brighten change" and "gaussian noise". These methods where chosen, because they did not change the object physical location on image. All new images where generated based on labelled images, that way the new images did not need to be relabelled. The original image json file was attached to artificially generated images. By using this method, a lot of marking process time saved. The original model showed an average of 73%, while the second model, which used data augmentation, improves the accuracy of the guess by an average of 16%, reaching an average of ~ 89%. The resulting network model was successful in identifying and localizing problems specific to oak wood.