An Improved Semantic Segmentation Model for Remote Sensing Images based on HRNet

Abstract

Deep learning has been greatly improved recently, and natural image processing based on deep learning has also been greatly improved. However, there are still great differences between natural images and remote sensing images, among which the biggest is that the size of the target in remote sensing images is greatly different, which requires the model to have a strong multi-scale processing ability. In order to meet this goal, we use HRNet with full multi-scale fusion capability to replace ResNet to process remote sensing images. HRNet fully integrates low-level detail features, middle-level structure features and high-level semantic features, which is very suitable for remote sensing images. The experimental results show that our method has been greatly improved.