In the paper, machine learning knowledge is transferred to wood processing industry. А universal Computer Vision system has great potential for the scalability. The labelling system based on convolutional neural networks was developed for quality inspection of the wood products to visually identify quality of surface and label by the smart camera in order to process and post-detect the visual non-compliance. The vision system consists of an inspection camera for capturing the image of the package and an image analysis software with Deep Machine Learning capabilities to identify undesirable, missing and damaged surfaces. Labelling defect detection increases the automation of the industry, making it less labour intensive, less costly and with improved efficiency of raw material. During the project, we proved that quality control inspection system with Deep Machine Learning technology developed for the Packaging/Food industry can be scaled up and using the same technology for the wood processing industry. Since the total number of defects for wood is quite large, given the variations in each of them it is difficult to build reliable system without huge dataset. Providing small datasets typically available for manufacturing systems usually results in weak models that lead to poor defect detection accuracy. We proved that using ML for wood defect detection and with a team of 6 workers it is possible to achieve a fully developed and sustainable defect recognition model within 3 months.