Page 110 - 《武汉大学学报(信息科学版)》2025年第9期
P. 110
第 50 卷 第 9 期 武 汉 大 学 学 报( 信 息 科 学 版 ) Vol.50 No.9
2025 年 9 月 Geomatics and Information Science of Wuhan University Sept. 2025
引文格式:虞欣,于家钰,郑肇葆,等 . 遥感影像场景识别的贝叶斯共轭批次归一化方法[J]. 武汉大学学报(信息科学版),2025,
50(9):1838-1847.DOI:10.13203/j.whugis20220632
Citation:YU Xin, YU Jiayu, ZHENG Zhaobao, et al. Scene Recognition of Remotely Sensed Images Based on Bayes Adjoint
Batch Normalization[J].Geomatics and Information Science of Wuhan University,2025,50(9):1838-1847.DOI:10.13203/j.whu⁃
gis20220632
遥感影像场景识别的贝叶斯共轭批次归一化方法
虞 欣 于家钰 郑肇葆 孟令奎 李林宜 2
1
1
2
2
1 北京石油化工学院人工智能研究院,北京,102617
2 武汉大学遥感信息工程学院,湖北 武汉,430079
摘 要:归一化方法作为特征预处理的关键部分,在浅学习和深度学习中都是至关重要的。针对批次归一化方法中存在
对批次样本容量依赖较大的问题,当前的优化思路主要是从样本信息的其他维度(如通道、层、时间等)来弥补批次样本
容量较小的不足。从贝叶斯理论的角度出发,通过将总体信息、先验信息和样本信息以科学、合理的融合方式来弥补批
次样本容量不足的缺陷,从而可以更加准确地估计样本均值和样本方差,使得归一化后的特征落入最佳的非饱和区域,
以便更好地反映整个特征空间的原始表征,进而深度学习模型可以达到最佳的特征表达能力。实验与分析表明,所提的
贝叶斯共轭批次归一化方法可行且有效,在 NWPU-RESISC45 数据集上,其分类精度比批次归一化方法高 5.64%。得
益于总体信息和先验信息,所提方法受样本容量的影响较小。
关键词:场景识别;批次归一化;共轭;贝叶斯
中图分类号:P237 文献标识码:A 收稿日期:2024⁃06⁃21
DOI:10.13203/j.whugis20220632 文章编号:1671⁃8860(2025)09⁃1838⁃10
Scene Recognition of Remotely Sensed Images Based on Bayes Adjoint
Batch Normalization
2
1
1
2
YU Xin YU Jiayu ZHENG Zhaobao MENG Lingkui LI Linyi 2
1 Academy of Artificial Intelligence, Beijing Institute of Petrochemical Technology, Beijing 102617, China
2 School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China
Abstract: Objective: Normalization method plays an important role in feature preprocessing phase not only
in conventional machine learning domain but also in contemporary deep learning domain. Batch normaliza⁃
tion (BN) is very successful, but its performance very depends on the sample size. Therefore, many re⁃
searchers try to improve it when the sample size is inadequate through adding the sample size merely in the
sample information space. Methods: This paper utilizes Bayes theory to integrate general information,
prior information and sample information, and to offset the inadequate sample information. In this way, the
mean value and the variance of sample can be estimated more precisely and more robust especially when the
sample size is small, and the normalized feature will better fall into non-saturating region of activation func⁃
tion, which enables deep learning model to better describe original feature space. Results: The Top-1 test
classification accuracy in NWPU-RESISC45 dataset is improved by 5.64% than that of BN method. More⁃
over, with the help of general information and prior information, the Bayes adjoint batch normalization
(BABN) method is not sensitive to the sample size. Conclusions: The experiment results show that the pro⁃
posed BABN method is feasible and effective, and performs better than BN method and other variants in
the scene recognition of remote sensing image.
Key words: scene recognition; batch normalization; adjoint; Bayes
基金项目:国家重点研发计划(2021YFB3900603);北京市科技新星计划(20250484800)。
第一作者:虞欣,博士,教授,主要研究方向为影像解译、人工智能和贝叶斯统计等。china_yuxin@163.com
通信作者:李林宜,博士,副教授。 lilinyi@whu.edu.cn

