摘要
语义分割是机器视觉中一项具有挑战性的任务,利用深度学习提高语义分割性能是当前研究的热点之一。针对木材缺陷图像语义分割问题,提出一种基于卷积神经网络(Convolutional Neural Network,CNN)的木材缺陷图像语义分割方法。首先,回顾CNN的几种典型的网络结构及其发展历程;然后,总结了图像语义分割方法的分类,并提出了改进的CNN图像语义分割方法;最后构建木材缺陷图像数据库,对模型进行训练和测试。基于TensorFlow与OpenCV的测试结果表明,设置合适的通道数和网络层数等参数,算法能够实现木材缺陷的图像分割。
Semantic segmentation is a challenging task in machine vision.Using deep learning to improve the performance of semantic segmentation is one of the hotspots in current research.Aiming at the problem of semantic segmentation of wood defect image,a method based on convolutional neural network(CNN)is proposed.Firstly,several typical network structures of CNN and their development history are reviewed;secondly,the classification of image semantic segmentation methods is summarized,and an improved CNN image semantic segmentation method is proposed.Finally,the wood defect image database is constructed to train and test the model.The test results based on TensorFlow and OpenCV show that the algorithm can achieve image segmentation of wood defects by setting appropriate parameters such as channel number and network layer number.
作者
严飞
程玉柱
YAN Fei;CHENG Yu-zhu(College of Mechanical and Electronic Engineering,Nanjing Forestry University,Nanjing 210037,China)
出处
《林业和草原机械》
2020年第6期52-56,共5页
Forestry and Grassland Machinery
基金
南京林业大学大学生创新工程项目(项目编号:2020NFUSPITP0118)
作者简介
严飞,男,本科,研究方向:木材阈值分割研究。
