An automatic remote sensing image shadow compensation method utilizing reflectance differences and transfer learning
Xingyu Shen, Yungang Cao, Baikai Sui, Shuang Zhang, Dejun Feng · GIScience & Remote Sensing · 2025
Shadows inevitably exist in remote sensing images (RSIs) and are characterized by lower resolution, mixed-pixel, and diverse background information. These characteristics pose a great challenge to the automatic compensation of shadows in RSIs. To address these issues, this paper proposes a training method using automatically simulated data to achieve real shadow compensation. In the process of data simulation, we take into comprehensive account the distribution, spatial extent, morphology, and other properties of shadows in RSIs, along with the reflectivity differences of ground features. And a lightweight shadow compensation framework is proposed to compensate shadows. Specifically, a Stair-step Hierarchical Context Aggregation Module is employed in the primary stage to integrate global and local texture information that is more important for shadow compensation. After that, a U-shaped end-to-end structure is used to perceive shadows and image features. Then, a Buffer-based mask attention Module is proposed to enhance the network’s attentiveness to both shadows and shadow boundaries. Finally, an additional high-resolution residual connection is utilized to get the final output. This connection effectively maintains the continuity of the image. This method was validated by two remote sensing image datasets with different resolutions and achieved favorable results.