摘要
针对合成孔径雷达(SAR)图像目标分类中扩展操作条件的重难点问题,提出了基于贝叶斯卷积神经网络与数据增强的SAR图像目标分类方法。该方法采用贝叶斯卷积神经网络获得更为可靠的分类网络,通过数据增强可为网络训练提供更为充足的样本数据,覆盖噪声干扰及部分遮挡等情形。实验结果表明,该方法在标准操作条件、噪声干扰及部分遮挡条件下,相比现有几类方法具有更强的有效性和稳健性。
Aiming at the key and hard problem in the target classification of synthetic aperture radar(SAR)images under extended operating conditions,a method based on Bayesian convolutional neural network and data augmentation was proposed.The method employed the Bayesian convolutional neural network to obtain more robust classification networks and produced more available samples via data augmentation,which covered the conditions of the standard operating condition,noise corruption,and partial occlusions.The experimental results validated that the proposed method achieved better effectiveness and robustness under the standard operating condition,noise corruption,and partial occlusion over several present algorithms.
作者
涂豫
TU Yu(Henan Polytechnic Institute,Nanyang 473000,China;Huazhong University of Science and Technology,Wuhan 430074,China)
出处
《探测与控制学报》
CSCD
北大核心
2020年第6期43-48,共6页
Journal of Detection & Control
关键词
合成孔径雷达
目标分类
贝叶斯卷积神经网络
数据增强
synthetic aperture radar
target classification
Bayesian convolutional neural network
data augmentation
