Fast quantitative estimation method of fracture cavity porosity based on convolution deep neural network

       Traditional interpretation of imaging logging data often involves manually importing various data types to calculate the porosity of fractures in the target area. This process becomes challenging due to the lack of gas and oil information in the raw data, especially when dealing with less-than-ideal raw data. The proposed method addresses this challenge by offering a rapid estimation approach for fracture porosity that reduces manual work and enhances process efficiency within an acceptable error limit.  

       The estimation method relies on path morphology [1] and convolutional neural networks for the extraction of fracture and cavity parameters. Initially, a path morphology method is applied to identify inclined fractures, followed by the use of a rotation jamming algorithm [2] to obtain rectangles with the minimum area in each cavity. These rectangles incorporate the angle of the rectangle and the lengths of its short and long sides. Parameters related to horizontal fractures, vertical fractures, and cavities are then utilized for the estimation of porosity.

      The original imaging logging conductivity is processed to distinguish inclined fractures from other fractures during the extraction process. Traditional binarization and denoising methods are not directly applied since cavities on basic binary images are also white. Thus, specific curves need to be extracted from the original conductivity images using a path morphology algorithm. On the other hand, convolutional neural networks (CNNs) are required for the identification of the shape of restored cracks due to the influence of cavities on traditional mathematical fitting processes. LeNet and AlexNet, among various CNN algorithms, are employed for this purpose. Specifically, the modified AlexNet algorithm adopts the maximum pooling method, Softmax function in the output layer, and the Adam optimizer in the learning process to improve efficiency and reduce memory occupation. The related parameters of cavities, horizontal fractures, and vertical fractures are calculated by the rotation jamming algorithm after the extraction of inclined fractures. Traditional Hough transform is considered time-consuming for evaluating a large number of cavities, leading to the adoption of an alternative approach—obtaining circumscribed rectangles with minimum area in a connected domain. This approach improves computation speed by focusing on directions coinciding with the long side of polygons, treating the long and short sides of rectangles as the major and minor axes of ellipses. In a conductivity image, cavities contain both convex and concave closures simultaneously, requiring the filling of concave ones to convex before applying the rotation jamming algorithm. The effective porosity parameters can be obtained using the developed programs.

      The method offers high efficiency and automation, extracting various types of fractures along with cavities within an acceptable error limit, providing valuable information for geologists in evaluating high-potential targets.

[1]Li et al. Estimating Porosity Spectrum of Fracture and Karst Cave from Conductivity Image by Morphological Filtering. JJU, 2017, 47(04): 1295-1307.

[2] Toussaint, G.T. A simple linear algorithm for intersecting convex polygons. The Visual Computer 1, 1985: 118–123. 

Figure1

Figure2

Figure3

Figure4