Remote sensing image data enhancement based on improved SinGAN

With the development of remote sensing technology, remote sensing images have been applied to a large number of fields such as remote sensing image recognition and segmentation detection. However, the problems of lack of remote sensing images, low quality and insufficient diversity hinder the performance improvement of remote sensing interpretation and other subsequent researches, and how to use a small amount of remote sensing images to generate a large number of datasets is an urgent problem at present. To address this problem, this paper combines a new pure convolutional network, ConvNeXt, with SinGAN network to build a remote sensing image data enhancement framework. Combined with ConvNeXt convolution network, the three image quality evaluation indexes of FID, SSIM and PSNR are increased by 5. 7%, 6. 2% and 8. 2%, respectively, on the remote sensing dataset NWPU-RESISC45 Dataset after combining ConvNeXt convolutional network for data enhancement. The quality and diversity of the data enhanced images based on the improved SinGAN remote sensing image data enhancement method are better than the SinGAN algorithm and the traditional image enhancement method, which can be used in remote sensing interpretation, change detection and other fields in practice.