Remote sensing image super-resolution reconstruction method, processing device and readable storage medium

Abstract

The invention discloses a remote sensing image super-resolution reconstruction method, a processing device and a readable storage medium. The remote sensing image super-resolution reconstruction method comprises the steps: preprocessing an image, constructing a generative adversarial network model, performing optimization, transfer learning and other processing on the generative adversarial network model, and finally obtaining a network model capable of outputting a super-resolution remote sensing image corresponding to an input low-resolution remote sensing image. According to the remote sensing image super-resolution reconstruction method, the generative adversarial network model is optimized by removing batch normalization of the residual module in the convolutional neural network; andthe generative adversarial network model is trained by using a transfer learning method, so that the problem that the model is difficult to train due to a small number of remote sensing images and lowquality is solved, and the performance index and the visual effect of a reconstruction result are improved while the memory consumption (about 40%) is reduced.

Description

technical field [0001] The present invention relates to the technical field of super-resolution reconstruction of optical remote sensing images, in particular to a remote sensing image super-resolution reconstruction method based on transfer learning and an optimized generative confrontation network model, a processing device and a readable storage medium. Background technique [0002] As an important means of obtaining ground object information, remote sensing technology is widely used in resource survey, land use, urban planning, crop yield estimation, national land census, building site selection, desertification monitoring, environmental protection, weather forecast, disaster monitoring, climate change and national defense military industry, etc. Both have achieved remarkable social and economic benefits and played an important role. The spatial resolution of remote sensing images is a key index to measure the quality of remote sensing images. Using high-resolution image...

AI Technical Summary

Problems solved by technology

[0008] 1. Unlike the super-resolution reconstruction requirements of optical natural images, the super-resolution reconstruction tasks of remote sensing images in actual situations often require large-scale reconstruction such as 4 times and 8 times. Super-resolution reconstruction networks based on ordinary neural networks are mostly It is only suitable for reconstruction tasks with small magnifications (2x, 3x), and it is difficult to obtain good reconstruction results in real application scenarios of remote sensing images;

[0009] 2. At present, most super-resolution reconstruction networks based on deep learning methods use batch normalization (BatchNormalization) for network training, but batch normalization is not suitable for super-resolution reconstruction tasks. It increases network parameters and reduces reconstruction accuracy. At the same time, it will increase the training burden of the neural network, reduce the "effective utilization rate" of the network, and remove the flexibility of the network after normalizing the features;

However, it is difficult for us to collect high-quality remote sensing images of so many scenes, targets, and conditions in reality, which restricts the application and development of super-resolution reconstruction tasks based on deep learning methods on remote sensing images.

Images