Non-paired remote sensing image thin cloud detection and removal method

Abstract

The invention provides an unpaired remote sensing image thin cloud detection and removal method, which is suitable for a small data set, and comprises the following steps: establishing a thin cloud imaging model, and designing a corresponding additive loss function to guide network training; the method comprises the following steps: constructing an unpaired thin cloud correction image small data set for a single-scene image, and providing a calculation method of reference thin cloud image data in combination with cloud wave band correlation; an alternately optimized multi-task network model is provided and can be adaptively applied to detection and removal of non-homogeneous thin cloud and mist, and the network can be suitable for data set expansion. According to the scheme provided by the invention, the requirement on data set construction is reduced, and the method can be effectively applied to thin cloud detection and removal of various remote sensing images, so that the thin cloud correction efficiency is improved.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, in particular to an unpaired thin cloud detection and removal method, which can be adaptively applied to non-homogeneous thin cloud correction and can achieve better performance on small data sets. Background technique [0002] The problem of thin cloud and fog has become an unavoidable problem in optical remote sensing images due to its wide distribution and high frequency. The thin cloud and fog areas in remote sensing images contain both part of the surface information and mixed cloud and fog information. The realization of thin cloud and fog correction for remote sensing images can improve the application potential of remote sensing data and the research accuracy of remote sensing applications. [0003] There are four main types of existing thin cloud and fog correction methods. 1) Correction method based on radiative transfer model; this method considers the physica...

AI Technical Summary

Problems solved by technology

The scarcity of paired data limits the use of existing mainstream deep learning methods on the problem of thin cloud correction; while the deep learning method based on unpaired data represented by Cycle-GAN is only suitable for large data sets, and the surface cover The correction effect of urban areas with frequent changes is not ideal

Images