应用BP神经网络识别内层表面细小缺陷的研究

内层表面细小凹坑的识别是超声无损评估的一个难点。本文利用人工神经网络对于信号的分类功能，建构和训练了一个BP神经网络，并用它对尺寸为1mm的圆锥形和半球形两种凹坑成功地进行了识别。研究表明，应用凹坑回波的DCT谱作为缺陷特片输入，可使BP神经网络的训练和缺陷识别既快捷又有效。

Ultrasonic identification of the shapes of tiny pits on an internal surface using BP neural networks

It is usually difficult to identify the tiny pits on an internal surface by ul trasonic nondestructive evaluation. Based on the classification ability of the artificial neural networks, a BP neural networks has been built, by which two kinds of pits are successfully identified. The research shows that using the DCT spectrum as the input of the networks enables the BP neural networks quick and effective in its training and identification.