A Laplacian based semi-supervised learning algorithm for radar target classification

We consider the classification problem of radar high range resolution profile with semi-supervised learning algorithm. Traditional practices are always supervised. They utilize the labeled data but discard the distribution information. In this paper, we take into consideration the unlabeled data and present a novel semi-supervised classification algorithm, called Laplacian Weighted Discriminant (LWD). Inspired by active learning, we first select the most representative points with Laplacian Transductie Optimal Design (LTOD). The sequence of selected points is used as the weight. Then the rate of average weighted distance to different kinds of labeled samples indicates the category of unlabeled samples. The experimental results have demonstrated the effectiveness of our proposed method.