| 摘要: |
| 为了提高合成孔径雷达（Synthetic Aperture Radar,SAR）目标识别性能，提出了结合二维内蕴模函数（Bidimensional Intrinsic Mode Function,BIMF)与贝叶斯多任务学习的SAR目标识别方法。采用二维经验模态分解获得SAR图像的多层次BIMF，从而更好地描述原始图像的细节信息。为了获得稳健的决策，采用贝叶斯多任务学习对原始SAR图像及其多层次的BIMF进行联合稀疏表示。最后，通过比较各个类别对于测试样本的重构误差判定目标类别。基于MSTAR数据集在多种条件下对提出方法进行了验证实验，结果证明了方法的有效性。 |
| 关键词: 合成孔径雷达 目标识别 二维内蕴模函数 贝叶斯多任务学习 |
| DOI： |
|
| 基金项目:宁夏高等学校科学研究项目(NGY2018-132,NGY2018-113)；宁夏师范学院2019校级重点科研项目(SFZD2019-19) |
|
| SAR Target Recognition via Combining Bidimensional Intrinsic Mode Function with Bayesian Multi-task Learning |
| ZHANG Hongwu,KANG Kai |
| (School of Mathematics and Computer Science,Ningxia Normal University,Guyuan 756000,China) |
| Abstract: |
| To improve the target recognition performance of synthetic aperture radar(SAR),a target recognition method is proposed which combines bidimensional intrinsic mode function(BIMF) with Bayesian multi-task learning.The bidimensional intrinsic model decomposition(BEMD) is employed to generate the multi-level BIMFs from the original image to better describe the detailed information of the original image.To obtain robust decisions,the Bayesian multi-task learning is used to jointly classify the original SAR image and its multi-level BIMFs.Finally,the target label is determined according to the reconstruction errors of different classes.Experiments on the MSTAR dataset under several conditions validate the effectiveness of the proposed method. |
| Key words: synthetic aperture radar target recognition bidimensional intrinsic mode function Bayesian multi-task learning |
《电讯技术》编辑部 