摘要
深度学习是当前机器学习的研究热点之一,针对木材表面死节缺陷图像,提出一种基于自动编码器(Autoencoder,AE)与长短期记忆网络(Long Short-Term Memory,LSTM)的深度学习图像分割方法。将RGB彩色图像转换成灰度图像,对灰度图像进行分块,同时将块变换成行向量,所有行向量组成矩阵并采用AE进行深度学习,通过设置多层深度学习结构,实现行向量维数约减。最后采用LSTM对约减后的死节和背景特征进行训练与测试并得到分类结果。试验结果表明,提出的算法的分割效果好,能很好地提取木材表面死节缺陷。
Deep learning is hot topic of current machine learning research field.Aiming at the defects of wood surface's dead knot,a surface crack image segmentation algorithm based on Autoencoder and Long Short-Term Memory (LSTM) is proposed.The RGB color image is converted into gray image,the gray image is divided into blocks and each block is transformed into a row vector.All the vectors are composed of matrix and the matrix is used as input data for deep learning by Autoencoder (AE).The dimension of row vector is reduced by setting layers of deep learning structure. Finally,LSTM is used to train the dead knot and background,and the classification results are obtained.Experimental results show that the proposed algorithm has a good segmentation effect and can effectively extract the dead-knot defects on the wood surface.
作者
程玉柱
李赵春
CHENG Yu-zhu;LI Zhao-chun(College of Mechanical and Electronic Engineering,Nanjing Forestry University,Nanjing,210037,China)
出处
《木材加工机械》
2018年第5期10-13,共4页
Wood Processing Machinery
基金
国家自然科学项目基金(51305207)
南京林业大学大学生创业项目(2016NFUSPIIP043)
