小波包-BP神经网络结合的热轧带钢缺陷识别

Defect Recognition of Hot Rolled Strip Based on Wavelet Packet and BP Network

基于热轧带钢的表面往往存在着很多缺陷,目前的识别方法存在着误识率高的问题,提出一种基于小波包分解的提取图像特征的方法,将提取的图像的能量特征向量输入BP神经网络分类器,对麻点、夹杂和结疤3种缺陷进行识别,仿真结果表明这种方法有着较高的识别率,并具有稳健的抗噪性和良好的扩展性。

As the current recognition methods have high misclassification rate for the defects on the surface of hot iolled steel strip, a new method of extracting image feature is proposed in view of wavelet packet decomposition. Through inputting the energy feature of the extracted image into BP neural network classifier, three defects of pit,inclusion and scar can be identified. Simulation results show that this method has high recognition rate,robust noise immunity and good scalability.