基于深度学习的木材缺陷图像检测方法  

Wood Defect Image Segmentation Based on Deep Learning

作　　者：程玉柱[1] 顾权 王众辉 李赵春[1] CHENG Yu-zhu;GU Quan;WANG Zhong-hui;LI Zhao-chun(College of Mechanical and Electronic Engineering,Nanjing Forestry University,Nanjing Jiangsu 210037,China)

机构地区：[1]南京林业大学机械电子工程学院,江苏南京210037

出　　处：《林业机械与木工设备》2018年第8期33-36,共4页Forestry Machinery & Woodworking Equipment

基　　金：南京林业大学大学生创新项目(2016NFUSPITP043)

摘　　要：针对木材活节、虫眼、死节等缺陷,提出一种深度学习的木材缺陷图像检测算法。首先构建训练数据库及测试数据库,同时设定卷积神经网络(CNN)的输入层、中间层、输出层等参数,并利用区域建议网络(RPN)反复训练CNN,然后利用训练好的CNN对测试图像进行检测,得到缺陷所在的矩形区域。将此区域作为初始分割范围,再利用CV模型进行图像精细分割。试验结果表明,提出的算法目标定位能力强,能很好地提取木材缺陷目标。In view of the defects of wood live knots,insect eyes and dead knots,a deep learning image detection algorithm for wood defects was proposed. First,the training database and the test database were constructed,and the parameters of the input layer,the middle layer and the output layer of the convolution neural network（CNN） were set,and the CNN was trained repeatedly by the regional recommended network（RPN）. Then the trained CNN was used to detect the test image,and the rectangle area of the wood defect was located. This region was used as the initial segmentation range,and the image was finely segmented using the CV model. The experimental results show that the proposed algorithm has strong target positioning ability and can well extract wood defect targets.

关 键 词：深度学习 卷积神经网络 区域建议网络 木材缺陷图像 CV 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]