文献类型：期刊文献

中文题名：木材树种计算机视觉识别技术发展与应用

英文题名：Advances and prospects of wood identification technology coupled with computer vision

作者：何拓[1] 刘守佳[1] 陆杨[1] 焦立超[1] 殷亚方[1]

第一作者：何拓

机构：[1]中国林业科学研究院木材工业研究所,中国林业科学研究院木材标本馆,北京100091

年份：2021

卷号：6

期号：3

起止页码：18-27

中文期刊名：林业工程学报

外文期刊名：Journal of Forestry Engineering

收录：CSTPCD;;北大核心:【北大核心2020】；CSCD:【CSCD_E2021_2022】；

基金：国家林业和草原局濒危物种管理行业规范项目(2019073024)。

语种：中文

中文关键词：计算机视觉识别技术;木材识别;深度学习;构造图像;特征自动化提取

外文关键词：computer vision identification technology;wood species identification;deep learning;anatomical image;automated feature extraction

分类号：S781

摘要：我国是全球林产品生产、贸易和消费第一大国,因此受到国际社会的广泛关注。在木材和木制品贸易流通环节经常出现以假乱真、以次充好的现象,为国际履约执法和林产品产业监管带来严峻挑战。基于木材解剖的传统木材树种识别方法,一般只能识别木材到“属”或“类”。近年来发展的DNA条形码、近红外光谱等木材树种识别新技术虽然可以实现木材“种”的识别,但难以在口岸、现场等多场景下对大批量样本进行自动精准识别。随着计算机技术的快速发展,计算机视觉识别技术可以从不同类别图像中提取关键特征,从而对图像进行分类,为木材树种分类带来新的途径。笔者首先介绍了基于图像采集、特征提取和树种分类的传统木材树种计算机视觉识别技术研究概况,然后从图像数据集构建、模型构建训练与测试以及系统开发等应用等方面介绍了基于深度学习的木材树种计算机视觉识别技术研究应用现状,并结合国内外研究进展对基于深度学习的计算机视觉识别技术在木材树种识别领域的应用进行了展望和提出建议,以期为木材树种自动精准识别研究提供新的思路。
Forests house over half of the world s wild plant and animal species,and produce valuable forest products for human.The over-exploitation of forest products poses huge threats to global biodiversity and ecosystem.China is one of the world s largest countries in the production,trade and consumption of forest products,and consequently,causes a broad concern all over the world.However,in the trade of wood and wood products,there are often mixtures of fakes and adulterants,which brings severe challenges to implementation and enforcement of international convention,and the supervision of the forest products industry.The critical step for implementation of international convention and national supervision of wood and wood products is definition identification of wood species.Traditional wood identification methods are based on wood anatomy,which generally can only identify wood to the“genus”or“class”level.Although other emerging wood identification technologies,i.e.DNA barcoding,near infrared spectroscopy can examine the“species”level of wood,it remains challenging to realize the automatic and accurate identification of a large number of samples at ports or on sites for filed screening of wood species,where the illegal activities of wood and wood products often happens.With the rapid development of computer technology in past decades,computer vision recognition technology could be deployed to extract key features from images of different classes for classification tasks,which brings an alternative approach for wood species identification.This study reviews the developments and applications of computer vision recognition technology for wood species identification.The research progress of traditional computer vision recognition technology for wood species identifications,including image acquisition,feature extraction and tree species classification,are reviewed.Additionally,the research advances and applications of computer vision recognition technology based on deep learning for wood species identification are further reviewed with multiple aspects,i.e.image data set establishment,model construction,training,and testing,as well as system development and applications.Conclusively,prospects and suggestions are proposed for the future application of computer vision recognition technology based on deep learning in the field of wood species identification,so as to provide scientific and technical supports for automated and accurate wood species identification.