基于神经网络/模糊逻辑系统的混合式特征提取器

特征提取是水声目标分类的关键环节之一，用以获取各类目标的一些可鉴别性特征。螺旋桨轴频、叶率和叶片数是目标的本质性、鉴别性特征，是分类的重要线索。本文提出了一个基于神经网络／模糊逻辑系统的混合式特征提取器，先利用侧抑制神经网络从目标辐射噪声的包络谱中提取线谱序列，再利用模糊逻辑系统从线谱序列中提取出螺旋桨轴频、叶率和叶片数。对目标数据的分析表明：该特征提取器能够为目标分类提取有效的鉴别性特征。

Hybrid Feature Extractor based on Artificial Neural Network and Fuzzy Logic Inference System

Feature extraction is one of key steps in underwater target classification system, to obtain some discriminatory information of various kinds of underwater targets. Shaft frequency, blade rate and number of blades of underwater targets are well known intrinsical discriminative features, and are important clues for classification . A hybrid feature extractor based on artificial neural network(ANN) and fuzzy logic inference system(FLIS) is proposed in the paper. At first, it extracts tone series using the ANN, which is suppressed in both sides of nodes, from the DEMON spectrum of target radiated noise, and then,extracts discriminatory information such as shaft frequency, blade rate and number of blades from the tone series using the FLIS. The results of analyzing real target-radiated noise data indicate that the feature extractor can efficiently extract these discriminatory features and has good performances.