Remote Sensing Image Segmentation Based on U-shaped Network with Atrous Spatial Pyramid

Semantic segmentation in remote sensing images has always been an important research direction of computer vision, and is widely used in land and resources related fields such as land mapping, hydrology and environmental testing, urban and rural construction. This paper focuses on remote sensing image segmentation problem, which is still challenging for current semantic segmentation networks. To address this problem, this paper takes the DeepLabv3+
^[1]
network as the baseline structure, and attaches a multi-level feature fusion module to well utilize the low-level information of the network. The proposed network performs better on multiple evaluations such as Precision, Recall, and Mean Absolute Error (MAE) compared to several state-of-the-art methods, and is able to extract details of the segmentation more precisely.