Application of a generative adversarial network for super-resolution of fracture images in LWD resistivity imaging logging

Objective The real-time Logging While Drilling (LWD) resistivity imaging technology visually presents the images of geological bodies near the wellbore, serving as an essential means for detailed evaluation of formation parameters and providing significant geological guidance in fractured reservoir drilling. However, constrained by the data transmission rate between the well and the surface, real-time data processing faces challenges of low image resolution, which adversely affects the qualitative identification of wellbore fractures and the quantitative evaluation of parameters. To address this issue, a generative adversarial network-based image super-resolution reconstruction method is proposed to enhance the clarity of fracture images.

Methods Firstly, a 23-layer deep generator network was constructed basing on the main framework of generative adversarial networks (GAN), integrating GAN with Residual Dense Blocks (RDB) and Residual Attention Mechanism (RAM). A dataset of high-resolution stored data images and low-resolution real-time data images was constructed using the actual measurements data from the RAB resistivity logging tool. Then, the constructed image dataset was trained using GAN, and network parameters with smaller errors and higher accuracy were obtained by optimizing the batch size and learning rate parameters. Finally, the trained network model was then utilized for super-resolution reconstruction of real-time data, achieving a resolution close to that of the stored data.

Results and Conclusions  It indicates that the proposed intelligent image super-resolution reconstruction method improves the peak signal-to-noise ratio (PSNR) and structural similarity index(SSIM) by 2.2 dB and 0.6, respectively, compared to traditional methods. Under fourfold down sampling, the algorithm can effectively reconstruct the image features of fractured geological bodies, dramatically enhancing the resolution of real-time LWD images. The method is of significant importance for improving the real-time geological guidance effect during drilling operations.