摘要
该文研究了训练样本不足的情况下利用卷积神经网络(Convolutional Neural Network,CNN)对合成孔径雷达(SAR)图像实现目标检测的问题。利用已有的完备数据集来辅助场景复杂且训练样本不足的数据集进行检测。首先用已有的完备数据集训练得到CNN分类模型,用于对候选区域提取网络和目标检测网络做参数初始化;然后利用完备数据集对训练数据集做扩充;最后通过"四步训练法"得到候选区域提取模型和目标检测模型。实测数据的实验结果证明,所提方法在SAR图像目标检测中可以获得较好的检测效果。
This paper studies the issue of SAR target detection with CNN when the training samples are insufficient. The existing complete dataset is employed to assist accomplishing target detection task, where the training samples are not enough and the scene is complicated. Firstly, the existing complete dataset with image-level annotations is used to pre-train a CNN classification model, which is utilized to initialize the region proposal network and detection network. Then, the training dataset is enlarged with the existing complete dataset. Finally, the region proposal model and detection model are obtained through the pragmatic “4-step training algorithm” with the augmented training dataset. The experimental results on the measured data demonstrate that the proposed method can improve the detection performance compared with the traditional detection methods.
作者
杜兰
刘彬
王燕
刘宏伟
代慧
DU Lan LIU Bin WANG Yan LIU Hongwei DAI Hui(National Laboratory of Radar Signal Processing, Xidian University, Xi＇an 710071, China Collaborative Innovation Center of Information Sensing and Understanding at Xidian University, Xi＇ an 710071, China)
出处
《电子与信息学报》
EI
CSCD
北大核心
2016年第12期3018-3025,共8页
Journal of Electronics & Information Technology
基金
国家自然科学基金(61271024
61322103
61525105)
高等学校博士学科点专项科研基金博导类基金(20130203110013)
陕西省自然科学基金(2015JZ016)~~
关键词
合成孔径雷达
目标检测
卷积神经网络
训练数据扩充
SAR
Target detection
Convolutional Neural Network （CNN）
Training data augmentation
