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

GeoCrack is the first large-scale open source annotated dataset of fracture traces from geological outcrops, enabling deep learning-based fracture segmentation, setting a new standard for natural fracture characterization datasets. GeoCrack contains images from photogrammetric surveys of fractured rock exposures across 11 sites in Europe and the Middle East, capturing diverse lithologies and tectonic settings. Each image was cleaned, normalized, and manually segmented, followed by a recursive annotation vetting process to ensure the quality and accuracy of the digitized fracture edges. The processed images and corresponding binary masks were divided into 224 × 224 patches, yielding 12,158 pairs. GeoCrack captures representive real-world challenges in fracture edge annotation, such as contrast variations between fracture traces and the host medium due to geological and geomorphological factors like aperture dilation, host rock composition, outcrop weathering, and groundwater staining. Physical occlusions like shadows and vegetation are also considered to minimize false positives. GeoCrack was validated using a U-Net implementation for fracture segmentation, achieving satisfactory IoU of 85%. GeoCrack holds strong potential to advance deep fracture segmentation in geological applications, effectively tackling the diverse challenges of real-world fracture identification.

show abstract

Identification of Polymer Flooding Flow Channels and Characterization of Oil Recovery Factor Based On U-Net

Cao,

Li,

Zhang

et al. 2024

Day 1 Mon, April 22, 2024

View full text Add to dashboard Cite

Image identification is a major means to achieve quantitative characterization of the microscopic oil displacement process. Traditional digital image processing techniques usually uses a series of pixel-based algorithms, which is difficult to achieve real-time processing of large-scale images. Deep learning methods have the characteristics of fast speed and high accuracy. This paper proposes a four-channel image segmentation method based on RGB color and rock particle mask. First, the micro model rock particle mask is divided together with the RGB component to form four-channel input data through image processing technology. Pixel-level training set labels are then created through traditional image processing techniques. Through the U-Net semantic segmentation network, the pixel-level oil and water identification and recovery factor calculation of the polymer microscopic oil displacement process were carried out. Combined with the pore distance transformation algorithm, the lower limit of pore utilization for different displacement media was clarified. The results show that U-Net can achieve accurate division of oil and water areas. Compared with conventional three-channel images, the improved four-channel image proposed in this paper has significantly improved the segmentation accuracy due to the addition of the constraints of the rock particle mask, and the global accuracy can be Up to 99%. Combining some post-processing methods, this paper found that polymer flooding increased the mobilization degree of small pores on the basis of water flooding and lowered the lower limit of pore mobilization from 25 μm to 16 μm. In microscopic experiments, the recovery factor was increased by 24.01%, finally achieving rapid and accurate quantitative characterization of the microscopic oil displacement process. The four-channel image method based on the U-Net semantic segmentation network and the improved rock particle mask proposed in this article has strong adaptability to the identification of flow channels in the microscopic oil displacement process. Quantitative characterization of the lower limit of pore movement and recovery degree during microscopic oil displacement provides a new method for microscopic image processing.

show abstract

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

Cited by 4 publications

References 67 publications

Real-time drone detection framework based on advanced texture feature extraction and pattern recognition model using GUI

Real-time drone detection framework based on advanced texture feature extraction and pattern recognition model using GUI

GeoCrack: A High-Resolution Dataset For Segmentation of Fracture Edges in Geological Outcrops

Identification of Polymer Flooding Flow Channels and Characterization of Oil Recovery Factor Based On U-Net

Contact Info

Product

Resources

About