探地雷达多目标识别方法的研究

与现有的机器学习算法相比，在有限样本的情况下，支撑矢量机具有更强的分类推广能力。该文在提出利用非线性映射进行探地雷达目标识别的基础上，将多目标识别支撑矢量机与探地雷达目标识别相结合，得到了基于一对一(One against one) 支撑矢量机的探地雷达多目标识别方法。所提方法包括基于一对一的探地雷达多目标识别方法、交叉验证的参数选取方法、多通道识别方法；并且和传统的神经网络识别方法进行对比分析。所提识别方法可以与各种目标特征选取方法相结合。对实测数据的对比处理表明所提方法优于传统探地雷达目标识别方法，所得结论对探地雷达目标识别的研究有指导意义。

Research on GPR Multi-object Recognition

With limited samples, SVM has stronger ability of generalization in comparison with machine learning algorithm. In this paper, the SVM is combined with the Ground Penetrating Radar(GPR) multi-object recognition, and a GPR multi-object recognition method is proposed based on the one against one SVM. The proposed method includes the GPR multi-object recognition method based on one against one SVM, the parameter selection method based on the cross-validation and the multichannel recognition method. The contrast analysis between the proposed method and the conventional neural network method is given. The proposed method can be combined with object-feature extraction methods. It is shown that the method is effective in the experimental analysis. The conclusion can direct the research on GPR object recognition.