Mapping wheel-ruts from timber harvesting operations using deep learning techniques in drone imagery

Wheel ruts, i.e. soil displacement caused by harvesting machines, should therefore be avoided or ameliorated. However, the mapping of wheel ruts that would be required in monitoring harvesting operations and planning amelioration measures is a tedious and time-consuming task. We used a deep-learning image-segmentation method (ResNet50 + UNet architecture) that was trained on drone imagery acquired shortly after harvests in Norway, where more than 160 km of wheel ruts were manually digitized. The leave-one-out cross-validation of 20 harvested sites resulted in F1-scores of 0.45-0.83 with an average of 0.67. The highest accuracy was obtained for severe wheel ruts (average user’s accuracy (UA) = 0.74), and the lowest accuracy was obtained for light wheel ruts (average UA = 0.62). Besides rut severity, the accuracy was also affected by the spatial resolution and noise present at the site. In combination with the ubiquitous availability of drones, the results of our study have the potential to greatly reduce the environmental impact of final felling operations by enabling the automated mapping of wheel ruts.