Semantic segmentation of ultra-high resolution remote sensing images based on fully convolutional neural networks

Semantic segmentation has been widely used in various fields of remote sensing; however, ultra-high resolution remote sensing images, due to their extremely high resolution, exhibit more spatial details and cannot be directly segmented using manual segmentation or traditional semantic segmentation methods. We applied fully convolutional deep learning-based neural networks to the semantic segmentation of ultra-high-resolution remote sensing images and at the same time designed an efficient image segmentation strategy to reduce the difficulty of training while ensuring the integrity of the dataset. We conducted experiments on the ultra-high resolution remote sensing dataset Potsdam and demonstrated that the semantic segmentation method based on fully convolutional neural networks can achieve 93.78% MIoU and 97.39% PA on the training set.