U-Net-based Semantic Edge Detection for Carbonate Reservoir Characterization

Summary This research introduces a novel U-Net-based deep learning model, specifically designed for accurate detection and segmentation of fracture edges in high-resolution geological outcrop images. Traditional methods for identifying fracture networks in rock formations often struggle with precision, especially in complex or noisy environments. The proposed model enhances the standard U-Net architecture by incorporating edge-aware skip connections and delineated convolutional layers, which are crucial for refining the detection of fine and intricate fracture edges. The integration of these innovations allows the model to effectively separate fractures from the surrounding rock matrix, even in challenging outcrop images where conventional methods often fail. The model has been trained and evaluated on the GeoCrack dataset, a high-resolution dataset of naturally fractured geological outcrops. Through this evaluation, the U-Net model demonstrated significant improvements in both accuracy and robustness, achieving an intersection over union (IoU) score of 85% in fracture segmentation tasks. This enhanced capability for edge detection accelerates fracture mapping workflows, offering substantial time savings and increased accuracy. The practical applications of this approach extend to fields such as carbonate reservoir characterization, slope stability analysis, and geothermal site assessment, making it a valuable tool for geoscientists and engineers working in these domains.