A Method Based on Generative Adversarial Networks for Completion of Blank Bands in Electric Logging Images

The electric logging image (ELI) is a valuable tool for revealing the underlying geological characteristics. However, due to the well structure and logging equipment limitations, ELIs can hardly cover 100% of the well perimeter and contain blank bands. Neural networks are widely adopted in image processing applications due to their excellent ability to capture image information. Therefore, based on convolutional neural networks (CNNs), this work proposes a method based on generative adversarial networks (GANs) for the completion of blank bands in ELIs. Specifically, it includes a generator network and two discriminator networks. The former is used to complete the blank bands to deceive the latter, and the latter is used to discriminate whether the ELIs are real or completed by the former. Optimization by adversarial training enables the generator network to generate more challenging adversarial samples, while the discriminator network can judge the authenticity of the input samples more accurately. In addition, to cope with ELIs with a large range of contextual information such as gravelly rock images with complex structures and textures, dilated convolutional layers are introduced into the generator network to increase the range of the network's receptive field and thus improve the model's performance. Ultimately, it is verified that the proposed method can effectively complete ELIs with blank bands.