自适应特征筛选的地雷目标AdaBoost分类器

The AdaBoost Classification of Land-mine Target with Adaptive Feature Selection

为解决前视地表穿透虚拟孔径雷达中地雷的分类问题,在传统AdaBoost算法的基础上,将特征选择作为弱分类器迭代的一部分,并将恒探测率下的虚警率作为特征选择的代价函数,提出一种基于弱分类器迭代及自适应特征选择的分类算法。通过实测数据验证,该分类算法适用于前视地表穿透虚拟孔径雷达中地雷与杂波的分类,同传统AdaBoost算法相比,分类性能有很大改善。

In order to solve the land-mine classification problem on a Forward-Looking Ground Penetrating Virtual Aperture Radar（FLGPVAR）,a new classification algorithm composed of weak classification iteration and adaptive feature selection is proposed.It is based on traditional AdaBoost algorithm,the feature selection is part of weak classification iterations,and the false alarm is treated as the cost function under constant detection rate.It is proved by real data that the method is applicable to the classification of land-mine and clutter on forward-looking ground penetrating virtual aperture Radar and the performance is better than traditional AdaBoost algorithm.