Research on the Design and Scene Adaptation of Remote Sensing Image Super-Resolution Models Based on CNN and Transformer Aggregation

In recent years, existing models in complex geographical scenarios (such as densely built urban areas and fragmented mountainous terrains), suffer from insufficient feature fusion accuracy and poor scene adaptability, leading to blurred edges and loss of details in super-resolved images. This limitation hinders their practical applications in fields like emergency disaster relief (e.g., rapid assessment of building damage in disaster areas) and agricultural yield estimation (e.g., precise identification of crop growth conditions). Many convolutional neural network (CNN)-based methods have made significant progress in single-image super- resolution tasks. However, CNNs only excel in local feature extraction while lacking in global feature modeling capabilities. To address this issue, we integrate CNNs(CNN), leveraging their local feature extraction strengths, with Transformers, which excel in global dependency modeling, to provide a novel approach for remote sensing image super-resolution. The CNN-Transformer aggregation mechanism enhances the model's adaptability to multi-scene remote sensing images, promoting the deep application of remote sensing super-resolution technology in practical operations (e.g., rapid emergency monitoring and precision agricultural yield estimation), thereby maximizing the value of remote sensing data.