Three-dimensional mine image super-resolution reconstruction method and system based on manifold discrimination

Abstract

The invention provides a manifold discrimination-based three-dimensional mine image super-resolution reconstruction method and a manifold discrimination-based three-dimensional mine image super-resolution reconstruction system, which are used for modeling the super-resolution problem of a single image and exploring an available HR space. The method and system is capable of: designing a natural manifold discriminator based on a CNN; providing a CNN structure based on irregular residual learning, namely, a fractal residual super-resolution (FRSR) network; and establishing a perception-oriented super-resolution NatSR network, so that real textures and natural details can be generated, and high visual quality is obtained.

Description

technical field [0001] The invention relates to the fields of coal mining and image processing, in particular to a method and system for super-resolution reconstruction of three-dimensional mine images based on manifold discrimination. Background technique [0002] The structure of mine underground tunnels is intricate, and parameters such as surrounding rock stress, water pressure, and water volume are directly related to the safe production of mines. At present, most of the casualties and property losses in mines are caused by underground safety accidents. With the development of computer science, virtual roaming technology has gradually emerged and has been applied to a certain extent. Applying virtual roaming technology to mine management, establishing a mine underground real scene model, and effectively monitoring underground equipment and environmental parameter information can make production personnel It is of great significance for mine production management and saf...

AI Technical Summary

Problems solved by technology

[0003] In the current state-of-the-art, many SISR models using CNN networks use distortion-oriented loss functions. Such models are difficult to restore real image texture content and details, which look blurry and have no good visual effects.

Restoring real texture and details is still a challenge in the field of image super-resolution, there are some works on this, such as SRGAN, EnhancedNet, SFT-GAN, however, these methods usually produce unrealistic details when generating these unrealistic details. required artifacts, the whole image always looks a little unnatural

In addition, the network trained with the objective loss function cannot reconstruct the real fine texture details

Images