Source-linked AI summary

A Remote Sensing Image Dataset for Cloud Removal

Daoyu Lin, Guangluan Xu, Xiaoke Wang, Yang Wang, Xian Sun, Kun Fu

arXiv:1901.00600v1cs.CV

TL;DR

Clouds and cloud shadows limit the use of optical remote sensing images, while deep-learning cloud removal lacks suitable training datasets. The paper introduces RICE, comprising two dataset components built from cloud, cloudless, mask, and quality imagery, and evaluates pix2pix with PSNR and SSIM. RICE is intended to promote deep-learning research on remote sensing image cloud removal.

  • Problem

    Clouds and cloud shadows limit optical remote sensing image use, while deep learning for cloud removal lacks suitable datasets for training neural networks.

  • Method

    The paper introduces RICE, including Google Earth cloud-cloudless image pairs and Landsat-derived cloud-removal data with quality information and temporally matched cloudless references.

  • Results

    Pix2pix was used to experiment on RICE, with RICE-II inputs formed from a cloud image and mask and accuracy measured using PSNR and SSIM.

  • Takeaways & Limitations

    RICE is intended to promote deep-learning development for remote sensing image cloud removal tasks.

Abstract

from arXiv · show

Cloud-based overlays are often present in optical remote sensing images, thus limiting the application of acquired data. Removing clouds is an indispensable pre-processing step in remote sensing image analysis. Deep learning has achieved great success in the field of remote sensing in recent years, including scene classification and change detection. However, deep learning is rarely applied in remote sensing image removal clouds. The reason is the lack of data sets for training neural networks. In order to solve this problem, this paper first proposed the Remote sensing Image Cloud rEmoving dataset (RICE). The proposed dataset consists of two parts: RICE1 contains 500 pairs of images, each pair has images with cloud and cloudless size of 512*512; RICE2 contains 450 sets of images, each set contains three 512*512 size images. , respectively, the reference picture without clouds, the picture of the cloud and the mask of its cloud. The dataset is freely available at \url{https://github.com/BUPTLdy/RICE_DATASET}.

1 Introduction

Optical remote sensing images are often degraded by clouds and shadows, limiting their use in downstream applications. The paper addresses the lack of datasets for deep-learning-based cloud removal by introducing the RICE benchmark dataset.

  • Clouds and cloud shadows reduce the utilization of optical remote sensing images and limit their further applications.
  • Deep learning has advanced remote sensing research and low-level vision tasks, including segmentation, detection, completion, defogging, denoising, and super-resolution.
  • RICE is a benchmark dataset proposed to promote deep learning for remote sensing image cloud removal.

2 RICE Dataset

RICE comprises two dataset components built from paired or temporally matched remote sensing imagery. Its sources and image products provide cloud, cloudless-reference, and quality information for cloud-removal research.

  • RICE-I: RICE-I contains 500 non-overlapping pairs of 512*512 cloud and cloudless images collected from Google Earth.
  • RICE-II: RICE-II is derived from Landsat 8 OLI/TIRS LandsatLook imagery and uses Natural Color and Quality images.
  • RICE-II: LandsatLook Natural Color images combine Landsat 8 OLI Bands 6, 5, and 4, while Quality images provide an 8-bit view of pixel quality.
  • RICE-II: Cloudless reference images for RICE-II were manually selected from the same location with acquisition times less than 15 days apart.

3 Experiments

The experiments use pix2pix to evaluate cloud removal on RICE, with RICE-II inputs formed by concatenating a cloud image and its mask. PSNR and SSIM are used for accuracy measurement.

  • Pix2pix was used to conduct experiments on the RICE dataset.
  • For RICE-II, pix2pix receives a concatenation of a cloud image and a cloud mask as input.
  • PSNR and SSIM were used to measure test-data accuracy.

4 Conclusion

The paper introduces RICE to support deep learning research on remote sensing image cloud removal. Future dataset expansion is planned across resolutions and scene categories.

  • RICE was introduced to promote deep learning development for remote sensing image cloud removal tasks.
  • Future expansion will add different resolutions and more scene categories.
Loading 1901.00600v1…