基于Hu不变矩和BP神经网络的木材缺陷检测

Detection of wood defects types based on Hu invariant moments and BP neural network

采用X射线作为检测手段,对木材进行无损检测,通过检测透过木材的射线强度来断定检测木材是否存在缺陷.对得到的木材缺陷进行图像处理,将木材缺陷图像转化为灰度图像,再把灰度图像转换为二值图像.根据经验选择相应的阈值,提取出清晰的木材缺陷边缘,把木材缺陷部位从背景中分离出来,完成木材缺陷图像分割.对Hu提出的区域不变矩进行扩展,得到一组新的描述形状特征的参数,这些参数具有平移、缩放和旋转不变性,并且具有较低的计算复杂性.将这些特征参数预处理后输入BP神经网络,对木材缺陷进行检测,检测准确率达到86%以上,试验结果表明此方法的可行性,为实现木材缺陷的自动检测提供了新的途径.

X-ray was adopted as a measure method for wood nondestructive testing.Wood defects w ere identified by testing X-ray transmitted intensity through the w ood.The detected defects w ere conducted by image processing.Wood defect images w ere first converted into grayscale images,and then into binary images.With the threshold values determined by some know n experience,the w ood defects w ere separated from the background and the clear w ood defects edge w as extracted.A group of parameters describing shape features w ere obtained by extending Hu invariant moments.Those parameters not only have translation invariance,scaling invariance and rotation invariance,but also have low er computational complexity.The feature parameters w ere input into BP（back propagation） neural netw ork after preprocessing,and then the w ood defects w ere recognized.The experimental results show that the recognition ratio is above 86%,indicating that this method is successful for detection and classification of w ood defects.This study offers a new method for automatic detection of w ood defects.