Research On Imaging Logging Image Restoration Method Based On Deep Learning

Imaging logging is a method of logging in which the borehole is measured by a unique logging instrument and the resulting data is converted into an image to better reproduce the true condition of the formation.However,in the process of logging,the imaging logging image is often distorted or partially missing due to collapsed well walls not recording complete data or individual pole plates not capturing data,and re-logging is not only expensive,but even impossible for completed boreholes.Therefore,how to accurately repair the missing portions of the imaging log image becomes a critical and worthwhile research problem.Traditional image restoration methods such as diffusion-based and texture-based methods lack understanding of image features and are unable to produce semantically structured and visually sound restoration results when restoring complex structural images such as imaging logging images.In recent years,with the in-depth study of deep learning,deep learning-based image restoration methods have achieved more outstanding restoration results.In this paper,we address the problem of partially missing imaging logging images and conduct research on deep learning-based image restoration of imaging logging images.The main contents are as follows:First,an improved UNet-based imaging logging image restoration method is proposed.The method uses the traditional UNet as the base network structure,extracts multi-scale features through hybrid cavity convolution,and introduces a spatial attention mechanism to design a bidirectional attention map module,so that the forward attention map module maintains mask updates and repairs missing regions,while the reverse attention map focuses on improving the quality of repaired regions.The experiments show that the method is applicable to the restoration of micro-resistivity imaging logging images,and has better improvement in semantic structure consistency and texture detail filling.Second,an improved GAN-based image restoration method for imaging logging images is proposed.The method addresses the problem of poor performance of the improved UNet’s restoration method in image edge-deficiency restoration,and designs a generative network and a discriminative network with reference to the game idea of generative adversarial networks.The generative network uses the full convolutional neural network FCN as the base structure,and designs and fuses the depth-separable convolutional residual block,Inception module,multi-scale feature extraction module and channel attention residual block,while the discriminative network designs the global discriminative network and local discriminative network.The algorithm is trained against the generative network to ensure that the restoration of the edge missing parts has semantic structure consistency with the global image,further improving the realism of the restored image.It is demonstrated that the algorithm is useful for improving the edge-defect repair of imaging logging images.Thirdly,the development of a deep learning-based image restoration system for imaging logs was completed.The system was developed based on the CIFLog platform and realised a number of functions such as imaging logging data import,imaging logging pre-processing,deep learning model training,deep learning model application and restoration result visualisation.The CIFLog platform has been tested and used with good overall results,and has a high value for use and promotion.