High-Resolution Cloud Detection Network

TL;DR

HR-cloud-Net tackles high-resolution cloud detection by preserving detailed cloud textures through a hierarchical high-resolution integration that combines a high-resolution representation module, layer-wise cascaded fusion, and a multi-resolution pyramid pooling mechanism. It introduces a multiview supervision strategy where a teacher view trained on normal images guides a student view trained on augmented images to improve generalization on limited data. Evaluations on CHLandSat-8, 38-cloud, and SPARCS demonstrate competitive or superior performance against state-of-the-art baselines, with improved edge fidelity and texture-based cloud delineation. The approach provides a practical, end-to-end solution for remote sensing cloud masking and is accompanied by publicly available code for replication and deployment.

Abstract

The complexity of clouds, particularly in terms of texture detail at high resolutions, has not been well explored by most existing cloud detection networks. This paper introduces the High-Resolution Cloud Detection Network (HR-cloud-Net), which utilizes a hierarchical high-resolution integration approach. HR-cloud-Net integrates a high-resolution representation module, layer-wise cascaded feature fusion module, and multi-resolution pyramid pooling module to effectively capture complex cloud features. This architecture preserves detailed cloud texture information while facilitating feature exchange across different resolutions, thereby enhancing overall performance in cloud detection. Additionally, a novel approach is introduced wherein a student view, trained on noisy augmented images, is supervised by a teacher view processing normal images. This setup enables the student to learn from cleaner supervisions provided by the teacher, leading to improved performance. Extensive evaluations on three optical satellite image cloud detection datasets validate the superior performance of HR-cloud-Net compared to existing methods.The source code is available at \url{https://github.com/kunzhan/HR-cloud-Net}.

Figures (7)

• Figure 1: High-resolution cloud detection network, HR-cloud-Net. (a) is the overall architecture of HR-cloud-Net, and (b) is high-resolution representation module details. Bottle Neck and Basic Block are derived from ResNethe2016deep.
• Figure 2: Multi-resolution pyramid pooling module.
• Figure 3: Multiview supervision of HR-cloud-Net. Aug denotes the data augmentation, and CE Loss is the cross-entropy loss function. The blue line represents the view of normal image feature, while the orange line represents the augmented image view.
• Figure 4: Visual comparison of HR-cloud-Net with other methods on the CHLandSat-8 dataset.
• Figure 5: Visual comparison of HR-cloud-Net with other methods on the 38-cloud dataset.