[Paper Review] A Remote Sensing Image Dataset for Cloud Removal

Daoyu Lin, Guangluan Xu+4 more|arXiv (Cornell University)|Jan 3, 2019

Advanced Image Fusion Techniques1 references73 citations

TL;DR

The paper introduces the RICE dataset for cloud removal in remote sensing images, comprising RICE-I (500 cloud/clear image pairs) and RICE-II (Landsat-derived cloudy/clear triplets), and provides baseline pix2pix results evaluated with PSNR and SSIM.

ABSTRACT

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}.

Motivation & Objective

• Motivate the need for cloud removal in optical remote sensing images and enable deep learning approaches.
• Provide a benchmark dataset (RICE) to support training and evaluation of cloud removal models.
• Offer baseline quantitative results to establish a reference for future methods.

Proposed method

• Construct RICE-I by obtaining cloud/clear pairs from Google Earth and cropping to 512x512 without overlap.
• Construct RICE-II from Landsat 8 LandsatLook images, aligning cloud-containing and cloudless references within 15 days.
• Use pix2pix as a baseline model for de-clouding, inputting a concatenation of cloud image and mask.
• Evaluate performance using PSNR and SSIM metrics.
• Report dataset details and qualitative considerations for future expansion.

Experimental results

Research questions

• RQ1 Can a paired dataset of cloud and cloudless remote sensing images support learning-based cloud removal?
• RQ2 What are baseline PSNR and SSIM performances for a conditional image translation model on RICE?
• RQ3 How do cloud removal results differ between the RICE-I and RICE-II subsets?

Key findings

• RICE-I achieves PSNR of 31.03 and SSIM of 0.91 in baseline evaluation.
• RICE-II achieves PSNR of 30.04 and SSIM of 0.80 in baseline evaluation.
• The dataset provides a public resource (GitHub) to promote deep learning-based cloud removal in remote sensing.
• Baseline experiments indicate the dataset’s utility for supervised cloud removal tasks.