Remote sensing image semantic segmentation method based on improved Deeplabv3+

Abstract

In recent years, with the continuous development of remote sensing technology and computer vision technology, the semantic segmentation of remote sensing images is of great significance in terms of earth observation, urban planning, military simulation, etc. This paper proposes a remote sensing image semantic segmentation method based on improved Deeplabv3+. Firstly, the backbone network is improved. Xception is selected to replace the traditional ResNe101 as the backbone network for the improved Deeplabv3+, and the network structure is deepened and depth separable. Optimization methods such as product replacement improve the segmentation efficiency; then, in order to improve the feature extraction effect of small targets in remote sensing images, the expansion rate of the cavity convolution in the ASSP module is optimized and adjusted. The experimental results show that the improved Deeplabv3+ algorithm has achieved good segmentation results on the data set, miou reached 91.23%, pixel accuracy reached 93.31%, and F1-score reached 89.2%, which is an increase of 2.4%,1.9% and 2.7% compared with the original Deeplabv3+. At the same time, compared with mainstream U-net and SegNet algorithms, this algorithm also has strong advantages in semantic segmentation of remote sensing images.