论文部分内容阅读
毫米波雷达通过发射宽带信号获得目标的结构信息。在毫米波一维距离像的识别中,有监督特征选择算法选择的特征子集相比无监督特征选择算法往往能更好的识别目标。然而在很多情况下,由于缺乏目标的类别信息,大量的样本无法得到充分利用。针对这个问题,本文应用标签重构算法,利用有限的已知标签样本集合构造训练集每个样本,得到样本的标签信息,并根据此将MCFS(multi-cluster feature selection)非监督特征选择算法推广为半监督的应用场合,得到基于标签重构的MCFS方法(LRMCFS)。实验结果表明,该算法与改进前算法以及同类算法相比,能够产生更好的识别效果,具有一定的优越性。
The millimeter-wave radar obtains the target structure information by transmitting a broadband signal. In the recognition of one-dimensional millimeter-wave range images, the subset of features selected by the supervised feature selection algorithm tend to be better at identifying targets than the unsupervised feature selection algorithms. However, in many cases, a large number of samples can not be fully utilized due to the lack of target category information. To solve this problem, this paper applies tag reconstruction algorithm to build each sample of training set with a limited set of known tag samples to get the tag information of the samples. Based on this, we extend the unsupervised feature selection algorithm of MCFS (multi-cluster feature selection) For semi-supervised applications, the MCFS method based on label reconstruction (LRMCFS) is obtained. Experimental results show that the proposed algorithm can produce better recognition results and has some advantages compared with the pre-improvement algorithm and similar algorithms.