Automatic Aircraft Shadow Removal from Remote Sensing Images Using Mask-ShadowGAN

Objects with shadow may cause a problem for image classification. For example, it can separate one object into many objects. It can also alter the size or shape of the object resulting in misclassification. In this paper, we focus on removing aircraft shadow from remote sensing images where the shadows occur on wings, bodies, and tails. Since it is very difficult to get shadow-free aircraft images and a shadow aircraft image of the same type for the training part, we adopted Mask-ShadowGAN for solving this issue. The benefit of the Mask-ShadowGAN algorithm is that, in the training part, the technique does not require the same images that have both shadow and shadow-free. In the experiment, we evaluated our proposed technique using RMSE and Jaccard similarity index for measurement. The experimental result shows that our technique shows promising results. We present both best and worst result based on sorted similarity index.