Abstract: To solve the problem of attenuation of the earth's surface reflectance in optical remote sensing images due to cloud cover, a thin cloud removal method based on the noise-adjusted principal component transform (NAPCT) model is proposed. Firstly, a NAPCT cloud removal model is constructed through noise estimation to realize the conversion of cloud images. Secondly, it corrects the turbid pixels on NAPC1 by using the cloud distribution information provided by the first component of NAPCT (NAPC1) and merging the cloud mask. Finally, the inverse transform is applied to the cloud covered area and mosaic with the original clear pixels to obtain a cloudless image. The performance of the method proposed in this paper is evaluated qualitatively and quantitatively using simulated and real Landsat 8 images. The experimental results show that the NAPCT method proposed in this paper provides better uniformity compared with other methods. The root square error and the peak signal-to-noise ratio have a better effect on removing clouds.

Key words: cloud mask, principal component transformation, cloud removal, signal-to-noise ratio, remote sensing image