基于贝叶斯集成算法的仿真SAR目标识别方法

发布时间：2018-04-24 03:04

本文选题：合成孔径雷达 + 贝叶斯集成算法　；参考：《中国电子科学研究院学报》2017年01期

【摘要】：SAR图像目标识别已经广泛的应用在众多领域,但是其对方位角较强的敏感性,会对识别结果造成较大的影响。特别地,同一类目标数据可能会因为不同的表现形式产生错误的识别的结果。而贝叶斯集成方法可以对不同的分类个体进行单独学习,并且因此获得不同的分类器,从而可以有效提高SAR图像识别准确率。在本文中,提出的基于一种集成算法的SAR目标识别方法,可以有效的解决上述问题。电磁仿真计算获得SAR图像包含多方位角,并从中进行特征提取。然后再将原始的训练样本集进行分组,每一个训练样本组可作为训练样本子集。这样就可以根据这些子集作为不同个体,从而得到不同的贝叶斯集成框架。并根据得到的框架完成识别,得到最终的识别结果。
[Abstract]:SAR image target recognition has been widely used in many fields, but its strong sensitivity to azimuth will have a great impact on the recognition results. In particular, the same class of target data may result in incorrect recognition due to different representations. Bayesian ensemble method can be used to study individual classifiers and obtain different classifiers, which can effectively improve the accuracy of SAR image recognition. In this paper, the proposed SAR target recognition method based on an ensemble algorithm can effectively solve the above problems. The electromagnetic simulation results show that the SAR images contain multiple azimuth angles and extract features from them. Then the original training sample set is grouped, and each training sample group can be used as a training sample subset. In this way, different Bayesian integration framework can be obtained by using these subsets as different individuals. Finally, the final recognition results are obtained according to the obtained framework.
【作者单位】：中国电子科学研究院;
【分类号】：TN957.52

【参考文献】