Cross-Modality Wood Log Tracing

Abstract

The proof of origin of logs is becoming increasingly important. In the context of Industry 4.0 and to prevent illegal logging there is an increased interest to track each individual log. In the near future more and more sawmills will be equipped with a computed tomography (CT) scanner. In order to establish wood log traceability from the forest to the sawmill this work investigates log recognition based on RGB log end images captured in the forest and CT log images captured in the sawmill. The advantage of that approach is that CT scanners are already applied in big saw mills to optimize the saw cut and so the logs only have to be recorded once more in the forest which saves time and cost. To bridge the domain shift between CT and RGB images, we apply widely known domain adaption approaches and present a novel filtering approach. Log recognition is done using a convolutional neural network (CNN) based method using the triplet loss for CNN training and a novel shape descriptor. The results (equal error rate of 13%) show that the recognition of logs using different imaging modalities (RGB and CT) is indeed feasible, despite the challenging experimental setup.