A Remote Sensing Image Dataset for Cloud Removal (1901.00600v1)

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}.

• The paper presents the RICE dataset, providing paired cloud-covered and cloud-free images to serve as a benchmark for cloud removal in remote sensing.
• The methodology involves using the pix2pix GAN framework, achieving PSNR values over 30 and SSIM scores up to 0.91.
• The dataset features two parts, RICE1 and RICE2, offering controlled images and enhanced cases with masks to drive future deep learning research.

Critical Analysis of "A Remote Sensing Image Dataset for Cloud Removal"

The paper "A Remote Sensing Image Dataset for Cloud Removal" addresses a significant gap in the field of optical remote sensing: the frequent obstruction of images by cloud cover, which hampers image analysis and subsequent applications in areas like Earth observation and environmental monitoring. Recognizing that deep learning has significantly advanced remote sensing tasks such as image segmentation and change detection, the authors identify a scarcity in suitable datasets for cloud removal tasks—a critical pre-processing step.

The authors introduce the Remote sensing Image Cloud rEmoving dataset (RICE), which consists of two distinctive parts, RICE1 and RICE2. RICE1 comprises 500 image pairs sourced from Google Earth, where each pair consists of a cloud-covered image and its corresponding cloudless version, both sized at 512x512 pixels. The dataset provides a controlled environment to test cloud removal models. RICE2, on the other hand, includes 450 sets of images derived from the Landsat 8 OLI/TIRS data. Each set contains three components: an image with clouds, a corresponding cloud mask, and a matching cloud-free reference image. This composition allows the utilization of additional features such as quality images and masks, which assist in refining cloud removal techniques.

The paper provides a preliminary evaluation of the dataset using the pix2pix framework, a well-regarded generative adversarial network (GAN)-based approach. The application of pix2pix to RICE1 and RICE2 yielded Peak Signal-to-Noise Ratio (PSNR) values of 31.03 and 30.04, and Structural Similarity Index (SSIM) scores of 0.91 and 0.80, respectively. These results demonstrate nominal yet promising effectiveness in cloud removal, reaffirming the viability of the dataset for further deep learning research.

The primary contribution of this paper lies in the release of RICE, which is anticipated to catalyze advancements in cloud removal methodologies within remote sensing. The availability of a benchmark dataset provides an essential resource for training and evaluating machine learning models, potentially enhancing the field's capability to interpret cloud-obscured optical imagery effectively.

The research suggests that future expansions of RICE could incorporate higher resolution images and additional scene categories to cater to diverse use cases. As deep learning continues to evolve, the availability of such comprehensive datasets is expected to facilitate the creation of more sophisticated models, refining the application of AI in remote sensing.

Overall, the creation of RICE marks a step towards addressing challenges in remote sensing analytics by provisioning crucial data for model training, encouraging methodological advancements, and potentially improving practical applications like climate monitoring and resource management.