Generative adversarial network-based remote sensing image thin cloud removal method

Abstract

The invention relates to a generative adversarial network-based remote sensing image thin cloud removal method. The method includes the following steps that: step 1, a thin cloud removal system modelis established; step 2, a network model is designed; step 3, an identification criterion function is constructed; step 4, the thin cloud removal system performs identification; and step 5, thin cloudsin a remote sensing image are removed. According to the method of the invention, the generative adversarial network is adopted to construct the model of an image thin cloud removal problem and perform system identification on the model, thereby realizing end-to-end thin cloud removal; and the constructed criterion function combines the errors of data distribution and reconstruction accuracy, so that the system can better learn the characteristics of data, thus, achieving the removal of the thin clouds. The method provided by the invention has a capacity to adaptively remove uneven thin clouds. With the method adopted, a restored image has high color and texture consistency.

Description

Technical field: [0001] The invention relates to a thin cloud removal method of a remote sensing image based on a generative confrontation network, and belongs to the application technical field of remote sensing image processing. Background technique: [0002] Remote sensing images provide rich information for earth science, meteorology, environmental monitoring, military monitoring and other tasks, and have become an increasingly important tool in modern measurement and control methods. However, there are large amounts of water vapor, ice crystals and fine dust condensation nuclei in the atmosphere, which gather in the form of clouds. However, electromagnetic waves are easily affected by clouds during the propagation process, causing the signal to be reflected or scattered, weakening the signal received by the sensor, affecting the imaging quality, resulting in the loss of information in the region of interest, and causing serious damage to the interpretation and interpret...

AI Technical Summary

Problems solved by technology

This discriminative model is a regression of existing data, and cannot learn the real distribution characteristics of the data well, so the generalization ability is not strong, and it does not perform well in complex scenes and various thin cloud situations.

Images