MRF-Net: An Infrared Remote Sensing Image Thin Cloud Removal Method with the Intra-inter Coherent Constraint

Qizhi Xu, Jiuchen Chen, Xinyu Yan, Wei Li

Abstract:

The usability of infrared remote sensing data is often compromised by thin cloud cover. To address this problem, we proposed the Multi-scale Residual Fusion Network (MRF-Net) to remove thin cloud from infrared remote sensing imagery.
First, to generate high-fidelity thin cloud, we developed a thin cloud simulation method based on Perlin noise and affine transformation.
Second, to accurately distinguish and remove thin cloud from infrared images, MRF-Net was proposed in this paper. The proposed model integrates the multi-scale feature fusion module to extract shallow features, the residual dense network module to extract in-depth features, the residual swin transformer module to capture global features, and attention mechanisms to focus on target information. Finally, we designed a combined loss function considering intra-block and inter-block constraints to ensure de-clouding consistency across different image blocks. The intra-block constraint was dedicated to removing thin cloud from each image block, while the inter-block constraint was designed to augment the consistency of cloud removal across different blocks. We have constructed a dataset that includes both simulated and real data to validate the efficiency of the proposed method. The experimental results demonstrated that the proposed method could eliminate thin cloud and outperform state-of-the-art methods. The source codes are available at \url{https://github.com/CastleChen339/MRF-Net}.