Research On Remote Sensing Image Enhancement Based On Multi-source Information Fusion

With the development of aerospace technology,more and more artificial satellites have entered space,and have provided tremendous help in the fields of social production,urban development,and military reconnaissance.However,remote sensing images are susceptible to interference from external factors such as atmosphere and electromagnetic waves,and traditional image enhancement methods based on single-source data cannot effectively utilize the multi-source characteristics of remote sensing images.Therefore,how to enhance the image quality through the complementary advantages of multi-source data and how to integrate image features have become a hot research issue in the field of image processing.This paper studies remote sensing image fusion technology.After analyzing the main limiting factors of remote sensing image quality,three image enhancement methods based on multi-source information fusion are proposed from three main issues.Firstly,in view of the noise existing in the hyperspectral image and the deficiency of the internal denoising method,an external denoising method based on the reference image is proposed.This method effectively utilizes the similarity information between the noise image and the clear band,With the proposed dual neural network,the features in the noise image and the reference image are extracted,and fused by the new activation function to finally obtain the noise-free image.Experiments show that this method has better denoising performance,especially when the image is much more noisy,the effect is far more than other algorithms.Secondly,this chapter studies the problem that night light data cannot have high temporal resolution and high spatial resolution at the same time,and proposes a spatiotemporal fusion method based on neural network.This method fuses the image at the pixel layer and the feature layer at the same time,and combining the advantages of two fusion level.In the experiment,Luojia-1 data and VNP46A1 data were fused,and the performance was not weaker than the traditional method.Finally,this paper studies the thin cloud removal method of remote sensing images,and proposes a thin cloud removal method based on the adversarial generation network and image fusion ideas.The adversarial generation network is used to simulate the real thin cloud information to make a simulation data set,and the historical cloudless image is merged into the cloudless network.The historical information is used to repair the image details.The experiment shows that this method can effectively remove the thin cloud in the remote sensing image.