一种应用于水声目标识别的隐层结构自适应网络

提出了一种隐层结构自适应学习的径向基函数网络（ＨＳＡＲＢＦ）水声目标分类器，该网络可在训练中自适应调整隐怪节点数和设置新隐节点的初始权值，从而使网络输入样本的分类特征能有效地映射到隐节点输出，克服了一般ＲＢＦ网隐层初始权值及隐节点难以确定的缺陷，经对实测水声信号的识别试验表明，该网络隐层有较强的特征划分能力，识别率高于一般ＲＢＦ网或ＢＰ网分类器。

一种应用于水声目标识别的隐层结构自适应网络

This paper proposes a hidden layer structure adaptive radial basis function (HSARBF) classifier. The number of hidden nodes can be adaptively adjusted and the initial weights jointed to additional nodes can be automatically assigned in the new method. Therefore, the class features of input samples can be effectively mapped to hidden layer. A few difficulties, as assigning initial weights and determining the number of hidden nodes in a traditional RBF classifier, can be well overcome. The effective division in feature space and high recognition rate of the classifier are demonstrated with experiments for real underwater acoustic signals.