TDFIF-Net: Two-dimensional feature interaction fusion network for digital core images

Abstract In the super-resolution(SR) reconstruction of digital core images for geological exploration, existing network models are limited to pass through a single dimension, and this limitation hinders the detailed structural analysis of the core, including pores and cracks, and makes it difficult to comprehensively capture the deep information of the core images. To solve the above problems, we propose a two-dimensional depth-based super-resolution reconstruction network for digital core images, which combines a bi-dimensional depth feature extraction unit (BDFM), discarding the traditional approach of feature extraction only from a single dimension, and instead exploring the deeper features of the core image from both spatial and channel dimensions simultaneously to take advantage of the complementary strengths of the two and achieve a more comprehensive feature interaction . In addition, we have designed a dual-domain synergistic enhancement (DDS) module, which effectively avoids the feature loss problem in the feature transfer process and accurately transfers feature information. Through extensive experimental validation on sandstone and carbonate datasets, our proposed model significantly outperforms the existing network models in terms of super-resolution reconstruction effect, with PSNRs of 30.3087 dB and 25.8508 dB on X4 scale for sandstone 2D and carbonate 2D datasets, respectively.This innovation not only provides accurate image support for geological exploration, but also contributes new deep learning solutions to the image super-resolution reconstruction field by contributing new deep learning solutions.The code is available at https://figshare.com/article/software/TDFIF-Net/256033614