基于仿真SAR和SVM分类器的目标识别技术研究

Research of Target Recognition Technique via Simulation SAR and SVM Classifier

SAR图像目标识别具有重要的军事应用价值,是目标自动识别领域中热点方向。传统的基于模板的识别方法随着识别种类的不断增多,其识别速度将不断降低;以基于支持向量机为代表的机器学习等新的识别技术,需要大量的真实SAR图像作为训练样本。针对以上问题,本文提出了一种基于仿真SAR和SVM分类器的目标识别方法。该方法首先通过电磁计算软件获得目标的大量仿真SAR数据构建训练样本集,然后利用样本集训练获得所需的识别分类器。本文通过实验验证了该方法对运输机、地面车辆分类识别的有效性,为SAR图像目标识别的应用推广提供了一条解决途径。

Target recognition of SAR image with important military application value is the key research on automatic target recognition. The computational velocities of conventional recognition method based on template match continually decrease with increasing target types. Novel recognition techniques via machine learning methods such as SVM et. al.,demand massive measured SAR image data. Aiming to solve the aforementioned problems,the target recognition technique via simulation SAR and SVM classifier is proposed. Firstly,the simulation SAR pattern is calculated by the software of electromagnetics to construct the training sample sets. Then,the recognition classifier is obtained based on the training samples.The experimental results of cargo-transport plane and vehicle demonstrate the effectiveness and robustness of the proposed method.