Enhancing Satellite Imagery: Time-Varying Neural Network Approach

Landsat images are essential in geoinformatics and earth observation. However, clouds frequently contaminate satellite images, making cloud removal increasingly necessary due to the growing demand for satellite imagery applications. In order to improve the quality of Landsat imagery, we suggest a novel Time-Varying Features Neural Network (TVFN-N) that consists of several modules: cloud detection, a learning module, feature fusion, and image reconstruction. Metrics including the Structural Similarity Index (SSIM), Root Mean Square Error (RMSE), and Peak Signal Noise Ratio (PSNR) are used to assess how effective the reconstructed images are as individuals. Our tests show that the suggested strategy is quantitatively successful, obtaining an RMSE of less than 0.0179. TVFN-N shows significant improvement in reconstructing Landsat images using multitemporal methods compared to older deep learning techniques. Additionally, we compared our approach using cutting-edge deep learning models and produced more accurate outcomes for jobs involving cloud removal.