基于证据K-NN改进算法的水声目标识别

UNDERWATER ACOUSTIC TARGET RECOGNITION BASED ON IMPROVED EVIDENTIAL K-NN ALGORITHM

证据分类算法已广泛应用于目标识别当中。针对传统证据K近邻算法在近邻证据组合规则上的局限,研究一种新的基于PCR5规则的证据K近邻改进算法(IEK-NN)。首先在总样本集中随机重复采样来构造多个训练子集;然后在各训练子集中,利用目标数据与其近邻的特征距离来构造基本置信指派;最后利用证据推理中的PCR5规则对近邻证据进行融合,并根据融合结果以及所建立的分类规则判断目标的类别属性。通过水声目标实测数据实验,将IEK-NN与传统的证据近邻分类算法进行对比分析,结果表明新算法能有效提高识别的准确率。

The evidence classification algorithm has been widely used in pattern recognition field. In view of the limitation of traditional evidential k-nearest neighbour classification algorithm in combination rule of nearest neighbour evidences, we proposed a PCR5 rule-based improved evidential k-nearest neighbour classification algorithm （IEK-NN）. First the new algorithm repeatedly samples from total sample set in random to construct a couple of training subsets. Then in each training subset, it uses the feature distance between target data and its nearest neighbour to determine the basic belief assignments. Finally the algorithm uses PCR5 rule of evidence reasoning to integrate the nearest neigh- bour evidences and according to the integration result and the classification rule established by it to judge the classification attribute of target. Through the experiment of measured data of underwater acoustic target, we make comparative analysis on IEK-NN and traditional evidential nearest neighbour classification algorithm. Result indicates that IEK-NN can effectively improve the recognition accuracy.