小波神经网络在缺陷数据压缩和信号重构中的应用  

Application of Wavelet Neural Network in Compression and Reconstruction of Flaw Signals

作　　者：康中尉[1] 罗飞路[1] 潘孟春[1] 陈棣湘[1] 

机构地区：[1]国防科学技术大学机电工程与自动化学院,长沙410073

出　　处：《无损检测》2005年第12期632-636,共5页Nondestructive Testing

摘　　要：首先介绍了小波神经网络理论,并在BP(反向传播)算法的神经网络学习方法的基础上,将小波神经网络应用到基于频率扫描技术的交变磁场无损检测系统中,通过神经网络提取相应的权重因子以及构成小波基的尺度参数和与之对应的平移参数实现缺陷有用数据的压缩;在缺陷数据重构中,利用上述特性参数并结合信号的特征值,对信号进行拟合,解决了缺陷场大量数据的保存问题以及缺陷识别模型神经网络学习样本资源不足的问题。Based on the theory of wavelet analysis and BP（ back propagation） algorithm learning, WNN（ wavelet neural network） was introduced to the nondestructive testing system based on alternative curtent field measurement （ACFM） technique. Data compression was accomplished by abstracting the characteristic parameters such as the weight coefficients, scale parameters and move parameters. On the other side, signal reconstruction was realized by combining the above characteristic parameters and the characteristic value of signal. The problems of the save of huge data and the shortage of the specimen for the neural network learning of flaw-recognition model were solved.

关 键 词：小波神经网络 数据压缩 数据重构 交变磁场测量 

分 类 号：TG115.28[金属学及工艺—物理冶金]