Automated mapping of bedrock-fracture traces from UAV-acquired images using U-Net convolutional neural networks

This contribution presents a novel U-Net convolutional neural network (CNN)-based workflow for automated mapping of bedrock fracture traces from 0,55 cm spatial resolution aerial photographs, acquired by unmanned aerial vehicles (UAV), over the Wiborg Rapakivi granite outcrops in the islands off the coast of the Loviisa Region in Southern Finland. The workflow comprised training a U-Net CNN using a small subset of photographs with manually traced fractures for optimizing the network parameters using the root mean squared propagation optimizer and sigmoidal focal loss function for semantic segmentation of input images and pixel-wise identification of fracture traces. The ridge detection algorithm was then applied to the U-Net prediction results, followed by vectorization of the fracture-traces pixels as vector polylines representing the traces of fractures. Both intensity values of the pixels and topological connectivity were used in the process of vectorization. Quantitatively the results were assessed using various accuracy assessment metrics. Qualitative evaluations of the results were implemented by comparisons of orientations and length-frequency distributions of automatically- and manuallymapped traces.The results show that the model has the class-balanced accuracy score 0.945, predicting 88.79% of the fracture-trace pixels. Bedrock outcrops with well-exposed surfaces present high true positive rates (>99%). The demonstration (test) site has the class-balanced accuracy score of 0.873 and 75% true positive rate. Additionally, the fracture trace networks replicate the orientation distributions of the manually digitized traces. The length distributions of automatic traces, however, differ with varying intensity from the manual trace length distributions. However, this study demonstrates that the workflow can be successfully applied to UAV-acquired images for fast and efficient automated mapping of bedrock-fracture traces during the initial phases of structural characterization of a region.

Automated summary

This contribution presents a novel U-Net convolutional neural network (CNN)-based workflow for automated mapping of bedrock fracture traces from aerial photographs acquired by unmanned aerial vehicles (UAV). The workflow includes training a U-Net CNN using a small subset of photographs with manually traced fractures, semantic segmentation of input images, pixel-wise identification of fracture traces, ridge detection algorithm and vectorization. The results show the effectiveness and accuracy of the workflow in automated mapping of bedrock fracture traces.