摘要
外来有害生物的鉴定一直是口岸检疫工作的重点,而对品种繁多的有害生物在种属层级进行准确识别则是检验检疫工作的难点,最新的计算机图像识别技术提供了解决这一问题的一种可行的途径。本文利用深度卷积网络的层次分类模型,对进口木材中经常截获的70种典型物种共9681张图片在科,属,种这3个分类层级上进行了识别鉴定,在1936张图片的测试中,模型在科,属,种上对70类有害生物的平均识别精度分别为97.71%,95.85%和86.92%。实验结果证明了模型对生物图像的学习能力,以及利用生物图像识别口岸有害生物的可行性。
The identification of exotic pest has always been the focus of quarantine work at ports.And identifying the various pests in species level or genera level accurately is a big challenge.Image recognition technology is a feasible way to solve this issue.In this paper,a hierarchical classification model based on CNN was used to identify 9681 pictures of 70 typical species which were intercepted from imported wood,at family level,genus level and species level.In the experiment on test dataset with 1936 images,the average classification accuracy at family,genus and species level of the model achieved 97.71%,95.85%and 86.92%respectively.The experimental results demonstrated the ability of our model to identify biological images and the feasibility of using biological images to identify port pests.
作者
孙佳佳
吕飞
雷晨曦
尚岩峰
熊惠霖
Sun Jiajia;Lü Fei;Lei Chenxi;Shang Yanfeng;Xiong Huilin(Taicang Customs House,Taicang 215400,China;School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University)
出处
《植物检疫》
北大核心
2020年第5期42-45,共4页
Plant Quarantine
基金
南京海关科技计划项目(2018KJ10)。
关键词
有害生物
图像识别
深度学习
卷积神经网络
鉴定
pest
image recognition
deep learning
convolutional neural network
identification