基于改进支持向量机的水声目标-杂波不平衡分类研究

The imbalanced classification of underwater acoustic target-clutter based on improved support vector machine

针对水声目标-杂波数据集在有限样本下的类不平衡特性导致代价敏感支持向量机难以逼近贝叶斯最优决策的问题,该文提出了一种基于能量统计方法的支持向量机(En-SVM)。该算法通过度量原始数据空间与有限样本空间特征函数之间的加权平方距离,量化少数类样本不完全采样过程中的信息损失,来补偿再生核希尔伯特空间中机器学习算法所需的少数类分类信息,增加少数类样本对决策的影响力。实验结果表明,En-SVM能够在保持高检测概率的同时获得较低虚警概率,即通过分类可以排除大量的杂波,性能优于标准支持向量机和代价敏感支持向量机,能够有效处理水声不平衡数据的分类问题,实现主动声呐信号处理中的杂波抑制。

This paper mainly proposed a novel algorithm En-SVM based on energy statistics method for the imbalance characteristics of acoustic target-clutter data sets resulted in cost sensitive support vector machines(CSSVM) didn’t approach Bayesian optimal decision in limited samples. This algorithm measured the weighted square distance between the original data space and the feature function of the limited sample space, quantifies the information loss in the incomplete sampling process of a few samples, so as to compensate for the minority class classification information required by the machine learning algorithm in the reproducing kernel Hilbert space, and increased the influence of the minority class samples on the decision-making. The experimental results show that the proposed algorithm can obtain a lower false alarm probability while maintaining a high detection probability, which means that a large amount of clutter can be eliminated by classification, and the performance is better than that of standard SVM and CS-SVM, which can effectively deal with the classification problem of underwater acoustic unbalanced data and realize clutter suppression of active sonar signal processing.