Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (8): 185-190.DOI: 10.3778/j.issn.1002-8331.2011-0021

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Research on Improved Semantic Segmentation of Remote Sensing

XIONG Fengguang, ZHANG Xin, HAN Xie, KUANG Liqun, LIU Huanle, JIA Jionghao   

  1. 1.School of Data Science and Technology, North University of China, Taiyuan 030051, China
    2.Simulation Equipment R&D Department, North Automatic Control Technology Institute, Taiyuan 030051, China
  • Online:2022-04-15 Published:2022-04-15



  1. 1.中北大学 计算机科学与技术学院,太原 030051
    2.北方自动控制研究所 仿真装备研发部,太原 030051

Abstract: In recent years, with the continuous development of remote sensing technology, remote sensing images have a great application prospect in the fields of urban planning, agricultural planning and military training. In this paper, a semantic segmentation method of remote sensing images based on improved Deeplav3 is proposed. By improving the single up-sampling layer, multi-layer up-sampling is carried out by using the residuals obtained from the backbone network to ensure the semantic integrity of the image on resolution; at the same time, the dilated rate of 4-layer dilated convolution in ASPP layer is modified to make the network have a better effect on small object segmentation. The experimental results show that, the proposed algorithm achieves 94.92% and 98.01% in mIou and pixel accuracy, which is 3.77 percentage points and 2.40 percentage points higher than the original algorithm. The proposed algorithm not only has a higher accuracy, but also has a better robustness to various terrain segmentation. Meanwhile, the proposed algorithm is also superior to state-of-art algorithm such as U-net, SegNet, HR-Net and DANet.

Key words: semantic segmentation, remote sensing, deep learning, up-sampling, dilated rate

摘要: 近年来,随着遥感技术的不断发展,遥感图像在城市规划、农业规划及军事训练等领域有着极大的应用前景,对遥感图像的语义分割研究也变得尤为重要。提出一种基于改进Deeplabv3的遥感图像语义分割方法,通过改进单一的上采样层,利用主干网络中得到的残差进行多层上采样,保证图像在分辨率上的语义完整;同时,修改ASPP层中4层膨胀卷积的膨胀率,使得网络对小物体分割有更好的效果。实验结果表明:改进的Deeplabv3语义分割算法在自制的数据集上mIou和像素准确率达到了94.92%和98.01%,较原算法分别提高了3.77个百分点和2.40个百分点,不仅拥有更高的准确性,而且对各类地形的分割有更好的鲁棒性;同时,提出的分割方法也优于U-net、SegNet、HR-Net和DANet等当前主流分割方法。

关键词: 语义分割, 遥感, 深度学习, 上采样, 膨胀率