Deep learning methods for fingerprint-based tracking in the wood industry

The wood industry places great emphasis on tracking methods, which allow a particular product to be recognised during the various steps of the production process. With this in mind, in this thesis, two neural networks will be analysed, which were developed to enhance the existing tracking system, thereby improving the recognition and timing of production facilities. In the realisation of the project, deep learning and fingerprinting methods were used. This procedure, which is able to exploit the biometric characteristics that make an item unique, has achieved excellent results in the field of facial and fingerprint recognition and, if applied to wood, would represent a great improvement for the industry. Indeed, it constitutes a non-invasive identification approach that allows the same board or log to be recognised at different points in the production chain. Therefore, the objective of this thesis is to test and further investigate the method in order to evaluate its effectiveness and future application. Two different convolutional neural networks (CNN) were used, one for the boards and one for the logs, both trained using triplet loss, which evaluate the similarity between two images in quickly and easily manner through the extraction of compact descriptors.