Abstract

Thick cloud removal for remote sensing images is an important and challenging task for researchers. Existed clouds removal methods always have some limitations with a large area of clouds or a long period between the cloudy image and the supplementary cloud-free image. In this paper, we proposed a deep-learning based framework for thick clouds removal. The method added prior spectral information into the model inputs and used deep convolutional neural networks (CNN) with dense connection and channel attention to reconstruct the cloudy areas. The loss function considered both spectral and structure similarity. We designed artificial and observed data experiments to show the performance of the network. Our method achieved the coefficient of determination (R<sup>2</sup>) of 0.976, structural similarity (SSIM) of 0.937 and root mean squared error (RMSE) of 0.016 in the artificial dataset and can generate reconstruction results with consistent spectral information and clear texture details, indicating that the proposed method is effective for cloud removal and data reconstruction.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call