【摘要】 水下目标信号的特征提取和分类是当今水声信号处理领域中存在的难题。随着人工神经网络技术的发展 ,众多的研究人员已致力于将人工神经网络应用于水下目标分类的研究中。本文给出了 FKCN( fuzzy Kohonen clustering network)算法 ,并将 FKCN应用于水下目标的分类问题中。实录海上无源声纳目标信号的分类实验验证了该算法的可行性。实验结果表明 :在大量的训练样本和测试样本下 ,FKCN提高了判别的灵活性 ,增加了判别的可信度 ,使系统的整体识别率提高约 2个百分点 ,较 KCN具有更好的分类效果。更多还原

【Abstract】 Because of the complexity of underwater environment, detecting and classifying underwater targets is a very difficult problem. Existing methods for classifying underwater targets are mostly based on artificial neural network. The deficiency of the existing methods is caused by neglect of some fuzzy information in target patterns. In this paper, we present a method based on fuzzy neural network for classifying underwater targets. We combine fuzzy c mean (FCM) algorithm with Kohonen clustering network (KCN) to form fuzzy Kohonen clustering network (FKCN), and extract the features of the targets by bispectrum to reduce the effects of non Gaussian noise. After introducing FCM algorithm (subsection 1.1), we give the learning algorithm of FKCN in subsection 1.2. Such an algorithm modifies its weights according to the cost function of FCM, so it overcomes the shortcoming of KCN′s dependence on input sequences. We use our method to classify three kinds of real underwater target signals from passive sonar, and satisfactory classifying results are obtained. The experiment results show that FKCN is better than KCN.更多还原