测绘学报 ›› 2019, Vol. 48 ›› Issue (6): 698-707.doi: 10.11947/j.AGCS.2019.20180434

• 摄影测量学与遥感 • 上一篇    下一篇

高分辨率光学遥感场景分类的深度度量学习方法

叶利华1,2, 王磊1, 张文文1, 李永刚2, 王赠凯2   

  1. 1. 同济大学电子与信息工程学院, 上海 201804;
    2. 嘉兴学院数理与信息工程学院, 浙江 嘉兴 314001
  • 收稿日期:2018-09-17 修回日期:2019-01-16 出版日期:2019-06-20 发布日期:2019-07-09
  • 作者简介:叶利华(1978-),男,博士生,讲师,研究方向为机器学习,深度学习,遥感影像解译。E-mail:9604ylh@tongji.edu.cn
  • 基金资助:
    国家重点研发计划(2017YFE0100900);浙江省自然科学基金(LY19F020017;LY18F020021)

Deep metric learning method for high resolution remote sensing image scene classification

YE Lihua1,2, WANG Lei1, ZHANG Wenwen1, LI Yonggang2, WANG Zengkai2   

  1. 1. College of Electronic and Information Engineering, Tongji University, Shanghai 201804, China;
    2. College of Mathematics, Physics and Information Engineering, Jiaxing University, Jiaxing 314001, China
  • Received:2018-09-17 Revised:2019-01-16 Online:2019-06-20 Published:2019-07-09
  • Supported by:
    The National Key Research and Develepment Program of China (No. 2017YFE0100900);The National Natural Science Foundation of Zhejiang Province of China (Nos. LY19F020017;LY18F020021)

摘要: 针对高分辨率光学遥感影像场景具有同类型内部差异大、不同类型间相似度高导致部分场景识别困难的问题,本文提出了一种深度度量学习方法。首先在深度学习模型的特征输出层上为每类预设聚类中心,其次基于欧氏距离方法设计均值中心度量损失项,最后联合交叉熵损失项以及权重与偏置正则项构成模型的损失函数。该方法的目标是在特征空间上使同类型特征聚集并扩大类型间的距离以提高分类准确率。试验结果表明,本文方法有效地提升了分类准确率。在RSSCN7、UC Merced和NWPU-RESISC45数据集上,与现有方法相比,分类准确率分别提高了1.46%、1.09%和2.51%。

关键词: 深度学习, 度量学习, 均值中心度量损失, 遥感影像, 场景分类

Abstract: Due to the similarity of intra-class and dissimilarity of inter-class of high-resolution remote sensing image scene, it is difficult to identify some image scene class. In this paper, a new classification approach for high-resolution remote sensing image scene is proposed based on deep learning and metric learning. Firstly, a clustering center of each class is preset on the output features of deep learning model. Secondly, the Euclidean distance method is used to calculate the average central metric loss. Finally, the final loss function consists of a central metric loss term, a cross entropy loss term, and a weight and bias term. The goal of this method is to improve the classification accuracy by forcing intra-class compactness and inter-class separability. The experimental results show that the proposed method significantly improves the classification accuracy. Compared with state-of-the-art results, the classification accuracy ratios on RSSCN7, UC Merced and NWPU-RESISC45 datasets are increased by 1.46%, 1.09% and 2.51%, respectively.

Key words: deep learning, metric learning, average center metric loss, remote sensing image, scene classification

中图分类号: