Inpainting Electrical Logging Images Based on Deep CNN with Attention Mechanisms

Microresistivity imaging logging is a crucial technique for evaluating the glutenite, carbonate and other complex reservoirs in oil exploration field. Because the size of the electrode pads in the electric imaging logging instrument can not match the borehole size well, it is hard to get full borehole covering images, which leads to many blank strips in the logging images. Thus it is necessary to fill the blank strips to ensure the accuracy of the subsequent processing and interpretation. The existing filling methods work well only in simple formation structures, but for images whose structure is very complex, for example, glutenite formation, the filled regions appear blurred and indistinct. To remedy the drawbacks, in this paper, we propose a machine learning method to inpaint the blank strips in electrical logging images based on deep convolutional neural network with the attention mechanisms embedded, which can capture a large number of low-level image statistical prior information. The attention mechanisms here refer to both the channel attention and the spatial attention. The channel attention focuses on the information to be attended in the intermediate feature maps while the spatial attention serves determining where to attend. Extensive experiments validate and confirm the effectiveness of the proposed filling method, which produce more compelling results than the existing filling approaches.