改进的Unet型木材缺陷图像分割方法

Improved Unet Model Wood Defect Image Segmentation Method

卷积神经网络(Convolutional Neural Network,CNN)是深度学习的最重要的网络之一,基于CNN的语义分割是当前研究的热点之一,Unet是在FCN基础上发展而来的卷积神经网络模型。针对木材缺陷图像分割问题,提出一种基于改进的Unet模型与像素阈值的木材缺陷图像分割方法。首先,在Unet的基本网络结构上,对网络的层数、通道数进行修改;然后,利用Unet实现网络训练,获得结构参数,最后用训练好的网络对图像进行测试,获取特征通道灰度图,并利用OTSU阈值算法对灰度图进行分割。结果表明,选择好的网络结构和阈值等参数,算法能够实现木材缺陷的图像分割,激活层通道灰度图分割效果优于卷积层。

Convolutional neural network(CNN)is one of the most important networks for deep learning.Semantic segmentation based on CNN is one of the hot topics in current research.Unet is a convolutional neural network model developed on the basis of FCN.In view of the problem of wood defect image segmentation,a wood defect image segmentation method based on improved Unet model and pixel threshold was proposed.Firstly,on the basic network structure of Unet,the number of layers and channels of the network were modified;then,the network training was realized using Unet to obtain the structural parameters.Finally,the trained network was used to test the image to obtain the characteristic channel gray image,and the OTSU threshold algorithm was used to segment the gray image.The results show that the algorithm can realize the image segmentation of wood defects by selecting the parameters such as network structure and threshold,and the segmentation effect of gray image of the activation layer channel is better than that of the convolution layer.