Remote Sensing Cloud Removal using a Combination of Spatial Attention and Edge Detection

High Resolution satellite images are of at most importance in the field of remote sensing. However, these images require quite a bit of preprocessing to ensure that the underlying landscape is not obstructed by any kind of unwanted noise. This paper addresses the problem of obstruction of remote sensing satellite data by clouds using a unique Generative Adversarial Network (GAN) model. Our proposed model Spatial Attention + Edges Generative Adverserial Network(SpA+Edges GAN) uses the spatial attention feature to focus on the regions of importance, namely the cloudy region during the reconstruction process. We combine this with the use of an edge filter that is used by the discriminator to compare the edges of the generated non-cloudy image and the cloud-free image. We also introduce a new loss function that forces the model to focus more on the cloudy region during the reconstruction process. We compare our model with other existing models on popular remote sensing datasets and also on a new dataset of our own using Peak signal to noise ratio (PSNR) and Structural Similarity index (SSIM). Through our experiments we show that combining the spatial attentive feature along with the edge filter provide promising results in removing clouds from remote sensing data.