A Progressive Remote Sensing Cloud Removal with Enhanced Global Context

Cloud removal in remote sensing imagery is crucial for applications in agriculture, environmental monitoring, and disaster assessment. However, existing methods often struggle to balance local details with global context, leading to significant information loss and degraded accuracy in downstream tasks. In this article, we propose a cascaded encoder-decoder network for progressive cloud removal that enhances global context understanding. Our architecture progressively refines feature representations to simultaneously capture global contextual information and preserve fine-grained local details. By integrating a triplet attention module and a fusion block, we strengthen hierarchical feature learning, enabling more accurate and visually consistent reconstructions of cloud-free images. Extensive experiments on the T-CLOUD, RICE1, and RICE2 datasets demonstrate that the proposed method achieves competitive and consistent performance in terms of PSNR, SSIM, and perceptual metrics. Furthermore, results from downstream scene classification tasks show that the restored images generated by our approach better retain discriminative semantic information compared to state-of-the-art methods.