Automatic modeling of high genus geological bodies using improved EdgeConnect and deep plug and play super resolution GAN

Rapidly and accurately constructing high genus geological models with serious intrusion and erosion can effectively help analyse engineering geological conditions. Modelling high genus geological bodies using a stratigraphic interface is complex and subjective. The method based on the voxel model has problems, such as redundancy of the voxel model, which makes it challenging to continue to exploit. Therefore, this study presents an automatic modelling method for high genus geological bodies. After converting the borehole model into a voxel image, accurate interpolation results were obtained using image completion and super-resolution algorithms. First, bilateral filtering and the Otsu method were used to improve the EdgeConnect completion algorithm and obtain preliminary interpolation results. Second, the residual-in-residual dense block (RRDB) and U-Net structure were used to optimise the network structure in the deep plug-and-play super-resolution generative adversarial network (DPSRGAN) to obtain a smoother and more accurate interpolation image. Finally, a voxel model simplification method was proposed to convert the redundant voxel model into a geological mesh model. Engineering practice has shown that the interpolation accuracy of this method is 88.4%. The number of model mesh surfaces is reduced by 91.7%. Compared with the geostatistical kriging interpolation, non-uniform rational B-spline (NURBS), radial basis function (RBF) and deep convolutional generative adversarial network (DCGAN) methods, the accuracy was improved by 19.2%, 3.7%, 11.0% and 20.9%, respectively, and the modelling time was shortened by 93.8% compared with manual modelling.