Remote sensing image super-resolution of open-pit mining area based on texture transfer

Remote sensing image is one of the main sources in open-pit mines production. Its spatial resolution affects the identification of the boundary of each scene in mining area, the interpretation of small ground objects and the location of control points. Remote sensing images play an important role in the open-pit production management and monitoring. Aiming at the problem of high spatial resolution image that meets the restrictions, such as cost and technological constraints, it is proposed to use a super-resolution technology to improve the spatial resolution of remote sensing images in open-pit mining areas. According to the obvious texture features of each scene in the open-pit mining area, the research used a deep learning texture transfer super-resolution method. Through an end-to-end deep learning model, the processes are as follows: input low-resolution images and corresponding reference images, divide them into several image patches, extract the features of image patches using feature extraction network, and compare low-resolution feature image blocks and reference features. The image patches texture similarity adaptively transfers the texture from reference images, and fuses the plurality of ex-changed texture feature maps into the generation network in the feature layers of various scales to construct the reconstructed images with rich texture details. At the same time, the ResNet34 with deeper network depth and smaller computing capacity is used to replace the VGG19, which further improves the feature extraction effect. The research used a self-made open-pit mining area dataset for experiments. Compared with the advanced image super-resolution methods, in terms of the influence of the reference image on the results, the results show that the improved method reconstruction effect will increase with the similarity between reference images and input images. Compared with other methods, the values of peak signal-to-noise ratio and structural similarity are better than the original methods, EDSR and SRGAN, and the visual perception is better. © 2019, Editorial Office of Journal of China Coal Society. All right reserved.