贝叶斯算法BP神经网络缺陷量化研究

为克服传统BP神经网络中网络训练速度慢、量化精度低、数据过度拟合、容易陷入局部极小点等缺点,该文将贝叶斯算法引入BP神经网络用于基于漏磁检测的缺陷量化,有效地控制网络模型的复杂度,利用不同尺寸的缺陷特征量训练网络,从而实现对缺陷长度、宽度、深度的量化,节约网络的训练时间,提高量化精度。

Quantification of slowly varying defect using BP neural network based on Bayesian algorithm

In order to overcome the disadvantages of traditional BP neural network such as slow training speed, low quantitative accuracy, data over fitting, easy to fall into local minima, this paper introduces the Bayesian algorithm to the BP neural network to quantify the defect through testing magnetic flux leakage. The BP neural network model is built to quantify the defect on the basis of the Bayesian algorithm. Bayesian reasoning is introduced to effectively control the complexity of the network model. And the defect features were used to train the network, so as to achieve the quantification of the length, width, depth of the defects. With this model, the training time of the network can be saved and the quantization accuracy can be improved as well.