Study On The Methods For Inpainting Blank Strips In Electrical Logging Images Based On Deep Learning

Electrical imaging logging is an effective technique for evaluating reservoirs.However,due to the characteristics of the borehole radius and the structure of the electrical imaging logging tool,the coverage rate can not reach 100% when the tool is opened,resulting in blank bands in the logging image,which affects the accurate interpretation of geological information.To ensure the accuracy of the subsequent processing and interpretation,it is necessary to fill these blank strips.Traditional blank strip filling methods work well in filling the sand-shale images with simple structure but are not effective for logging images with complex structure,such as glutenite and carbonate.Recently,deep learning(DL),especially convolutional neural networks(CNN)has found to be able to capture the abstract information of images at a high level and improve the representation ability of deep image features,which makes it been successfully applied in the field of image processing compared with traditional methods.To this end,this paper focuses mainly on the study of blank strips inpainting methods in complex electrical logging images based on deep learning.The main work of this paper is as follows:1.First,the digital image restoration methods are analyzed and summarized,especially the implementation principles of traditional image restoration methods as well as the deep learningbased methods are investigated in-depth.Then the characteristics,advantages and disadvantages of various methods are compared,which lays a solid theoretical foundation for the subsequent work.2.A blank strip filling model of electrical logging images based on a deep encoder-decoder network is proposed.Compared with the regular deep learning filling model,aiming to make the network lay stress on the features that are more important to the filling task,an attention mechanism is introduced to the model to extract salient image features.Meanwhile,in order to improve the modeling ability of geometric transformations for the deep learning neural network,the basic convolution in the network is replaced with deformable convolution,which will be beneficial to the restoration the edge contour information of complex images.Experiments on the glutenite data set with complex structure show that the image filled by the proposed model is clearer and more consistent with the context content.Meanwhile,the contour edge of the glutenite is better restored.The filling effect has been significantly improved compared with the traditional methods and the basic deep neural network methods.3.A blank strips inpainting model of electrical logging images based on generative adversarial network is proposed.By using the generative confrontation network trained on large-scale natural images,the model can obtain richer depth prior information than a single log image,thereby realizing the blank filling of the electrical logging images of the glutenite with complex structure.Experimental results show that the proposed method can better restore the edge structure and texture details of the glutenite,which gets more realistic results.Furthermore,the filling traces are hardly be distinguished by human eyes and the consistency of the image is much better.