Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (7): 207-213.DOI: 10.3778/j.issn.1002-8331.1806-0024

Previous Articles     Next Articles

U-Net Based Semantic Segmentation Method for High Resolution Remote Sensing Image

SU Jianmin, YANG Lanxin, JING Weipeng   

  1. College of Information and Computer Engineering, Northeast Forestry University, Harbin 150040, China
  • Online:2019-04-01 Published:2019-04-15

基于U-Net的高分辨率遥感图像语义分割方法

苏健民,杨岚心,景维鹏   

  1. 东北林业大学 信息与计算机工程学院,哈尔滨 150040

Abstract: Image segmentation is an important base-part of remote sensing interpretation. High resolution remote sensing image contains complex object information, but the applications of traditional segmentation methods are greatly limited. The segmentation method, represented by the deep convolution neural network, has made a breakthrough in many fields. Aiming at the problem of high resolution remote sensing image segmentation, this paper proposes a deep convolution neural network based on U-Net, which achieves the end to end pixel level semantic segmentation. It expands the original dataset, trains a binary classification model for every class of objects, and then combines the prediction subgraphs to generate the final semantic segmentation image, which has helped us get 94% training accuracy and 90% test accuracy on the dataset of AI classification and recognition contest of CCF satellite images. The experimental results show that the network not only has good generalization ability but also can be used in practical engineering with high segmentation accuracy.

Key words: remote sensing image, semantic segmentation, convolutional neural network, U-Net, ensemble learning

摘要: 图像分割是遥感解译的重要基础环节,高分辨率遥感图像中包含复杂的地物目标信息,传统分割方法应用受到极大限制,以深度卷积神经网络为代表的分割方法在诸多领域取得了突破进展。针对高分辨遥感图像分割问题,提出一种基于U-Net改进的深度卷积神经网络,实现了端到端的像素级语义分割。对原始数据集做了扩充,对每一类地物目标训练一个二分类模型,随后将各预测子图组合生成最终语义分割图像。采用了集成学习策略来提高分割精度,在“CCF卫星影像的AI分类与识别竞赛”数据集上取得了94%的训练准确率和90%的测试准确率。实验结果表明,该网络在拥有较高分割准确率的同时还具有良好的泛化能力,能够用于实际工程。

关键词: 遥感图像, 语义分割, 卷积神经网络, U-Net, 集成学习