| 引用本文: | 常沛，夏勇，李玉景，吴涛. 基于CNN的SAR车辆目标检测[J]. 雷达科学与技术, 2019, 17(2): 220-224.[点击复制] |
| CHANG Pei, XIA Yong, LI Yujing, WU Tao. SAR Vehicle Target Detection Based on CNN[J]. Radar Science and Technology, 2019, 17(2): 220-224.[点击复制] |
|
| 摘要: |
| 传统的SAR目标检测算法容易受到复杂背景的干扰，因此利用被广泛应用于图像目标检测和识别领域的Faster-RCNN方法，对复杂背景下的SAR图像进行车辆目标检测实验。在对样本数据进行预处理后对车辆真实位置进行标记，采用可视化的深度学习客户端对样本进行裁剪和旋转，扩充样本数据库。利用已充分训练的模型权重对ZF和VGG-16网络进行预训练，再利用扩充的数据集进行训练和验证，并使用包含MiniSAR数据的测试集进行测试。实验证明，ZF网络和VGG-16的检测效果类似，但是ZF网络因为网络层数更少因而检测耗时更短。 |
| 关键词: 合成孔径雷达 卷积神经网络 数据扩充 目标检测和识别 |
| DOI：DOI:10.3969/j.issn.1672-2337.2019.02.016 |
| 分类号:TN958；TP753 |
| 基金项目: |
|
| SAR Vehicle Target Detection Based on CNN |
|
CHANG Pei, XIA Yong, LI Yujing, WU Tao
|
|
1.The 38th Research Institute of China Electronics Technology Group Corporation, Hefei 230088, China;2.Key Laboratory of Aperture Array and Space Application, Hefei 230088, China
|
| Abstract: |
| Traditional SAR target detect algorithm is easily disturbed by complex image scenes, thus Faster-CNN method, which has been widely used in image target detection and recognition field, is applied to accomplish SAR image vehicle target detection in complex background. After pre-processing the sample data, the ground truth location of the vehicle is labeled, and the visual deep-learning client is used to crop and rotate the sample to expand the sample database. The ZF and VGG-16 networks are pre-trained with well-trained model weights, then trained and verified by use of extended datasets, and finally tested with test-sets containing MiniSAR data. Experiments show that the detection effects of ZF network and VGG-16 network are similar, but ZF network takes less time because of the fewer number of network layers. |
| Key words: synthetic aperture radar (SAR) convolutional neural network (CNN) data augmentation target detection and recognition |
