一种基于组合核函数支持向量机的水下目标小波特征提取与识别方法

Hybrid Kernel Function Support Vector Machine Based Method for the Extraction and Recognition of Wavelet Feature of Underwater Targets

准确的模式识别要求提取出的特征尽可能反映分类本质的特征.本文利用同态分析理论对水下声信号进行预处理,从最终接收到经过噪声干扰的目标信号中复原出能反映目标传输特性的原始信号,并在此基础上对信号进行离散小波变换,提取小波变换系数在不同区间上的尺度—过零密度、尺度—平均幅度特征,最终利用组合核函数支持向量机对提取出的特征进行分类识别.实验表明,提取出的特征能反映目标类别特点,该方法能对水下目标进行有效的识别.

Precisely pattern recognition require that the extracted feature should reflect the essential character of classification as far as possible, In this paper, the underwater acoustic signal is preprocessed according to the theory of homomorphic analysis, the original signal which reflects the target＇ s transport character therefore can be reconstructed among the interfered noise signal. Based on this step, discrete wavelet transform is performed to extract features such as scale-zero cross density in multi interval and scale-magnitude from the transformed coefficients, Then a hybrid kernel function support vector machine is designed to recognize different targets. Experiments show that this method can achieve good performance in the underwater target recognition field.