【摘要】 在无损检测信号处理和特征构造的基础上 ,用神经网络对缺陷进行识别 ,然后运用模糊积分对多个神经网络的分类结果进行融合。并以胶接结构典型缺陷的超声波检测与识别为例 ,给出了一些实验结果。结果表明用模糊积分集成后的正确分类率比单独网络的正确分类率有较大提高更多还原

【Abstract】 In this paper, defects for NDT are classified with neural network which uses features extracted from signal processing. Then classification results from networks are fused with fuzzy integral. The implementation of the method and some experiment results are given. Classification results with the method show that correct classification probability is significantly great than individual neural network.更多还原