FIB-SEM super-resolution algorithm based on generative adversarial network

Abstract

The invention relates to the technical field of image resolution processing, and discloses an FIB-SEM super-resolution algorithm based on a generative adversarial network, and the algorithm comprises a discriminator module which carries out the training of a low-resolution image, and obtains a high-resolution image with a corresponding multiplying power, and carries out the discrimination of the high-resolution image; the generator module is used for converting the low-resolution image into a preset high-resolution image; according to the method, an FIB-SEM high-resolution image generating an adversarial network is reconstructed by using a low-resolution image based on a deep learning algorithm, and the authenticity of the resolution of the reconstructed image is judged through a discriminator module; meanwhile, a loss function combination is introduced in the training process of the low-resolution image and the original image to serve as an optimization target, so that the converted high-resolution image is more real, and the method has remarkable advantages in the aspects of solving the ill-conditioned inversion problem caused by information loss of the low-resolution image, anti-noise performance and the like; furthermore, the reconstructed high-resolution image shows more microstructures such as pores, throats and cracks, and the precision of a traditional experiment is effectively expanded.

Description

technical field [0001] The invention belongs to the field of image resolution processing, and in particular relates to a FIB-SEM super-resolution algorithm based on a generative confrontation network. Background technique [0002] In recent years, China is vigorously developing cleaner unconventional oil and gas, mainly shale oil and gas, with low pollution and low carbon emissions, which not only ensures national energy security, but also contributes to the realization of the dual-carbon goal. Compared with conventional oil and gas reservoirs, unconventional oil and gas reservoirs such as shale oil and gas have very tight reservoir rock formations, and the accurate analysis of their microscopic pore structure characteristics in the process of exploration and development is extremely demanding. Most of the pores in tight reservoirs such as shale are nano-scale. At present, focused ion beam scanning (FIB-SEM) is often used for 3D imaging to extract pores. One-time imaging by...

AI Technical Summary

Problems solved by technology

[0003] With the expansion of data sets and the continuous deepening of training model depth, the super-resolution technology of a single image has been developed rapidly, but at present, many super-resolution algorithms applied to FIB-SEM technology scanning images do not perform well

Many training sets are trained only with a single, fixed low-resolution and corresponding high-resolution images. This level of degradation cannot satisfy the real image scene evolution scanned by FIB-SEM technology.

On the other hand, the degradation model is more of an image model based on linear interpolation, but only after nonlinear processing such as tone correction and lossy compression, will the image reconstructed by the model appear dashed lines and artifacts

The application of high-resolution graph convolutional neural network and self-attention module in image super-resolution tasks has been successful, but most of the research is only based on the Euclidean distance between image pixels after imaging as a measure of image imaging effect. The cost function, this cost function does not intuitively reflect the human visual perception

Images