iWood: 基于卷积神经网络的濒危珍贵树种木材自动识别系统

doi:10.11707/j.1001-7488.20210915

摘要/Abstract

摘要：

目的: 构建基于卷积神经网络的木材识别系统，实现木材树种在多场景条件下的自动精准识别，为我国提升CITES履约执法能力、加强林产品产业链监管以及保障木材安全提供科技支撑。方法: 采集15种黄檀属和11种紫檀属木材标本横切面构造特征图像，建立图像数据集Rosewood-26；构建AlexNet、VGG16、DenseNet-121和ResNet-50共4种卷积神经网络模型，基于ImageNet图像数据集对模型进行迁移学习，采用Rosewood-26图像数据集训练、测试和比较模型，优选识别性能较好的卷积神经网络模型，并进行木材树种分类；在此基础上，构建包含15种黄檀属和11种紫檀属树种的木材自动识别系统iWood，利用市场木材样品对系统进行应用测试和评价。结果: 在构建的4种卷积神经网络模型中，ResNet-50模型表现出最高的识别精度（98.33%）、最少的权重数量和较低的模型复杂性，适用于木材树种准确快速识别；ResNet-50模型对9种黄檀属和3种紫檀属木材的识别精度达100%，并可成功鉴别构造特征极其相似的檀香紫檀和染料紫檀；基于ResNet-50模型构建的木材自动识别系统iWood，在"属"和"种"水平的识别精度分别为91.8%和77.3%。结论: 基于卷积神经网络的木材识别系统iWood适用于海关执法、木材贸易和质量监督检验等多场景下的木材自动精准识别，能够为我国提升CITES履约执法能力、加强林产品产业链监管以及保障木材安全提供科技支撑。

关键词: iWood, 卷积神经网络, 木材构造特征, 图像数据集, 木材自动识别, 识别精度

Abstract:

Objective: An image recognition method based on convolutional neural networks(CNNs) for endangered and precious tree species was developed in this study to reach the automatic and accurate identification of timbers at the species level in multi-scene scenarios. Method: Images representing wood anatomical features were collected from transverse section of authentic wood specimens for 15 Dalbergia and 11 Pterocarpus species, and an image data set of Rosewood-26 was established. Four CNN models, i.e. AlexNet, VGG16, DenseNet-121 and ResNet-50 were constructed and pre-trained with ImageNet for transfer learning. And then the image data set of Rosewood-26 was deployed to re-train and test these models, which were comparatively analyzed and evaluated to obatin an optimal one for wood identification. A wood identification system was developed for identifying those samples collected from timber market for application test. Result: Among the four CNN models, ResNet-50 showed the highest identification accuracy(98.33%) and lower model complexity, which is preferable in the context of accurate and rapid wood identification. The ResNet-50 model achieved 100% accuracy when identifying 9 Dalbergia and 3 Pterocarpus species, and successfully discriminated P. santalinus from its look-alike species, P. tinctorius. The automated wood identification system based on ResNet-50 model exhibited an identification accuracy of 91.8% at the genus level and 77.3% at the species level. Conclusion: The wood identification system, iWood, developed in this study using CNNs is applicable in broad fields, i.e. customs enforcement, timber trade and quality inspection, and can reach automatic and accurate identification of wood species. This study will provide scientific support for promoting regulation of forest product industry chain, enhancing CITES enforcement capabilities and protecting forest species diversity.

Key words: iWood, convolutional neural networks, wood anatomical features, image dataset, automated identification, identification accuracy

中图分类号:

S781.1

何拓,刘守佳,陆杨,张永刚,焦立超,殷亚方. iWood: 基于卷积神经网络的濒危珍贵树种木材自动识别系统[J]. 林业科学, 2021, 57(9): 152-159.

Tuo He,Shoujia Liu,Yang Lu,Yonggang Zhang,Lichao Jiao,Yafang Yin. iWood: An Automated Wood Identification System for Endangered and Precious Tree Species Using Convolutional Neural Networks[J]. Scientia Silvae Sinicae, 2021, 57(9): 152-159.

图/表 5

表1

图1

图2

表2

图3

参考文献 0

	邓旭冉, 闵少波, 徐静远, 等. 深度细粒度图像识别研究综述. 南京信息工程大学学报: 自然科学版, 2019, 11 (6): 625- 637.
	Deng X R , Min S B , Xu J Y , et al. A survey of deep fine-grained visual categorization. Journal of Nanjing University of Information Science and Technology: Natural Science Edition, 2019, 11 (6): 625- 637.
	何拓, 高瑞清, 焦立超, 等. 世界木材标本馆现状与发展建议. 木材工业, 2020, 34 (3): 40- 43.
	He T , Gao R Q , Jiao L C , et al. Present status of global leading xylaria and suggestions for xylaria development in China. China Wood Industry, 2020, 34 (3): 40- 43.
	何拓, 焦立超, 郭娟, 等. 木材信息学: 发展、应用与展望. 木材科学与技术, 2021a, 35 (4): 15- 24.
	He T , Jiao L C , Guo J , et al. Wood informatics: history of development, application, and prospective trend. Chinese Journal of Wood Science and Technology, 2021a, 35 (4): 15- 24.
	何拓, 刘守佳, 陆杨, 等. 木材树种计算机视觉识别技术发展与应用. 林业工程学报, 2021b, 6 (3): 18- 27.
	He T , Liu S J , Lu Y , et al. Advances and prospects of wood identification technology coupled with computer vision. Journal of Forestry Engineering, 2020, 6 (3): 18- 27.
	胡明越. 2019. 基于深度学习的树种识别算法研究. 杭州: 浙江农林大学硕士学位论文.
	Hu M Y. 2019. The study of tree species identification based on deep learning method. Hangzhou: MS thesis of Zhejiang A&F Univeristy. [in Chinese]
	姜笑梅, 殷亚方, 刘波. 木材树种识别技术现状、发展与展望. 木材工业, 2010, 24 (4): 36- 39.
	Jiang X M , Yin Y F , Liu B . Current status, development and prospect of wood identification technology. China Wood Industry, 2010, 24 (4): 36- 39.
	焦立超, 何拓, 尹江苹, 等. 第18届CITES缔约方大会木材树种管制变化. 木材工业, 2019, 33 (6): 24- 28.
	Jiao L C , He T , Yin J P , et al. Changes of CITES-listed timber species in Cop 18th. China Wood Industry, 2019, 33 (6): 24- 28.
	李桂兰. 2007. 亚洲主要商品木材构造与识别研究. 南宁: 广西大学硕士学位论文.
	Li G L. 2007. Research on the structure and identification of wood from Asia. Nanning: MS thesis of Guangxi University. [in Chinese]
	刘子豪, 祁亨年, 张广群, 等. 基于横切面微观构造图像的木材识别方法. 林业科学, 2013, 49 (11): 116- 121.
	Liu Z H , Qi H N , Zhang G Q , et al. Wood identification method based on microstructure images in cross-section. Scientia Silvae Sinicae, 2013, 49 (11): 116- 121.
	梅萍, 夏兆鹏, 张耀丽, 等. 檀香紫檀和染料紫檀快速微损鉴别方法. 福建农林大学学报: 自然科学版, 2017, 46 (2): 154- 158.
	Mei P , Xia Z P , Zhang Y L , et al. Micro-loss method of identifying Pterocarpus santalinus and Pterocarpus tinctoricus. Journal of Fujian Agriculture and Forestry University: Natural Science Edition, 2017, 46 (2): 154- 158.
	王登举. 全球林产品贸易现状与特点. 国际木业, 2019, (3): 49- 53.
	Wang D J . Status quo and characteristics of global trade of forest products. International Wood Industry, 2019, (3): 49- 53.
	殷亚方, 何拓, 焦立超, 等. 2019. 一种木材识别方法及系统: ZL201810830841. 3.
	Yin Y F, He T, Jiao L C, et al. 2019. A method and system for wood identification: ZL201810830841. 3. [in Chinese]
	Brancalion P H S , Almeida D R A , Vidal E , et al. Fake legal logging in the Brazilian Amazon. Science Advances, 2018, 4 (8): eaat1192. doi: 10.1126/sciadv.aat1192
	Deklerck V , Mortier T , Goeders N , et al. A protocol for automated timber species identification using metabolome profiling. Wood Science and Technology, 2019, 53, 953- 965. doi: 10.1007/s00226-021-01302-9
	He K , Zhang X , Ren S , et al. Delving deep into rectifiers: surpassing human-level performance on ImageNet classification. Proceedings of the IEEE International Conference on Computer Vision, 2015, 1026- 1034.
	He T , Lu Y , Jiao L , et al. Developing deep learning models to automate rosewood tree species identification for CITES designation and implementation. Holzforschung, 2020a, 74 (12): 1123- 1133.
	He T , Macro J , Soares R , et al. Machine learning models with quantitative wood anatomy data can discriminate between Swietenia macrophylla and Swietenia mahagoni. Forests, 2020b, 11 (1): 36.
	Hinton G E, Srivastava N, Krizhevsky A, et al. 2012. Improving neural networks by preventing co-adaptation of feature detectors. arXiv: 1207. 0580.
	Iandola F, Moskewicz M, Karayev S, et al. 2014. DenseNet: implementing efficient ConvNet descriptor pyramids. arXiv: 1404. 1869.
	Jiao L , Yu M , Wiedenhoeft A C , et al. DNA barcode authentication and library development for the wood of six commercial Pterocarpus species: the critical role of xylarium specimens. Scientific Reports, 2018, 8, 1945.
	Kanayama H , Ma T , Tsuchikawa S , et al. Cognitive spectroscopy for wood species identification: near infrared hyperspectral imaging combined with convolutional neural networks. Analyst, 2019, 144, 6438- 6446.
	Krizhevsky A , Sutskever I , Hinton G E . ImageNet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems, 2012, 1097- 1105.
	LeCun Y , Bengio Y , Hinton G . Deep learning. Nature, 2015, 521, 436- 444.
	Lewis S L , Edwards D P , Galbraith D . Increasing human dominance of tropical forests. Science, 2015, 349, 827- 832.
	Nair V , Hinton G E . Rectified linear units improve restricted boltzmann machines. Proceedings of the 27th International Conference on Machine Learning, 2010, 807- 814.
	Pan S J , Yang Q . A survey on transfer learning. IEEE Transactions on Knowledge and Data Engineering, 2010, 22 (10): 1345- 1359.
	Pennington J, Schoenholz S, Ganguli S. 2017. Resurrecting the sigmoid in deep learning through dynamical isometry: theory and practice. Advances in Neural Information Processing Systems 30(NIPS 2017).
	Ravindran P , Costa A , Soares R , et al. Classification of CITES-listed and other neotropical meliaceae wood images using convolutional neural networks. Plant Methods, 2018, 14, 25.
	Ravindran P, Ebanyenle E, Ebeheakey A A, et al. 2019. Image based identification of Ghanaian timbers using the XyloTron: opportunities, risks and challenges. arXiv: 1912. 00296.
	Ravindran P , Wiedenhoeft A C . Comparison of two forensic wood identification technologies for ten Meliaceae woods: computer vision versus mass spectrometry. Wood Science and Technology, 2020, 54, 1139- 1150. doi: 10.1007/s00226-020-01178-1
	Simonyan K, Zisserman A. 2014. Very deep convolutional networks for large-scale image recognition. arXiv: 1409. 1556. .
	Wiedenhoeft A C , Simeone J , Smith A , et al. Fraud and misrepresentation in retail forest products exceeds U. S. forensic wood science capacity. PLoS ONE, 2019, 14 (7): e0219917.
	Zhang M , Zhao G , Guo J , et al. A GC-MS protocol for separating endangered and non-endangered pterocarpus wood species. Molecules, 2019a, 24 (4): 799.
	Zhang M , Zhao G , Liu B , et al. Wood discrimination analyses of Pterocarpus tinctorius and endangered Pterocarpus santalinus using DART-FTICR-MS coupled with multivariate statistics. IAWA Journal, 2019b, 40 (1): 58- 74.

标签 Lable	木材树种 Wood species	木材样本数量/图像采集数量Number of wood samples/collected images
标签 Lable	木材树种 Wood species	BCTw	CAFw	MADw、SJRw	SPSFw	YZ
0	赛州黄檀Dalbergia cearensis	7/56	0	5/137	2/15	5/50
1	交趾黄檀Dalbergia cochinchinensis	0	1/18	5/102	1/5	5/50
2	绒毛黄檀Dalbergia frutescens var.tomentosa	7/43	0	8/307	1/15	5/50
3	海南黄檀Dalbergia hainanensis	0	1/29	1/46	0	0
4	黄檀Dalbergia hupeana	0	7/118	2/21	0	0
5	阔叶黄檀Dalbergia latifolia	3/17	2/12	15/626	1/7	5/50
6	东非黑黄檀Dalbergia melanoxylon	2	0	9/189	0	5/50
7	巴西黑黄檀Dalbergia nigra	26/160	1/6	21/565	8/93	5/50
8	降香黄檀Dalbergia odorifera	0	4/87	0	0	5/50
9	奥氏黄檀Dalbergia oliveri	0	2/66	4/211	0	5/50
10	微凹黄檀Dalbergia retusa	0	0	16/556	0	5/50
11	印度黄檀Dalbergia sissoo	3/33	0	18/662	0	5/50
12	亚马逊黄檀Dalbergia spruceana	6/53	0	3/105	3/18	5/50
13	伯利兹黄檀Dalbergia stevensonii	0	0	11/407	0	5/50
14	危地马拉黄檀Dalbergia tucurensis	0	0	12/478	0	5/50
15	安哥拉紫檀Pterocarpus angolensis	2/12	2/67	15/492	0	5/50
16	安达曼紫檀Pterocarpus dalbergioides	0	0	12/347	0	5/50
17	刺猬紫檀Pterocarpus erinaceus	0	5/62	4/116	0	5/50
18	印度紫檀Pterocarpus indicus	1/9	2/66	25/1067	1/9	5/50
19	大果紫檀Pterocarpus macrocarpus	1/7	3/40	13/470	1/7	5/50
20	囊状紫檀Pterocarpus marsupium	2/23	0	13/461	0	5/50
21	药用紫檀Pterocarpus officinalis	1/11	0	20/661	0	0
22	罗氏紫檀Pterocarpus rohrii	1/8	0	9/368	0	0
23	檀香紫檀Pterocarpus santalinus	0	1/26	4/132	0	5/50
24	非洲紫檀Pterocarpus soyauxii	3/15	4/48	6/204	5/58	5/50
25	染料紫檀Pterocarpus tinctorius	0	1/30	5/154	0	5/50

网络模型 CNN models	模型深度 Depth of models	权重数量 No. of weights(×10⁶)	训练时间 Training time/h	模型精度 Model accuracy(%)
AlexNet	5	61.0	3.58	93.85
VGG16	16	138.4	11.62	98.00
DenseNet-121	121	27.2	26.13	96.00
ResNet-50	50	25.5	8.05	98.33

[1]	赵子宇,杨霄霞,郭慧,葛浙东,周玉成. 基于卷积神经网络模型的木材宏、微观辨识方法[J]. 林业科学, 2021, 57(6): 134-143.
[2]	谢锦莹,丁丽霞,王志辉,刘丽娟. 基于FCN与面向对象的滨海湿地植被分类[J]. 林业科学, 2020, 56(8): 98-106.
[3]	郭颖,李增元,陈尔学,张旭,赵磊,陈艳,王雅慧. 一种改进的高空间分辨率遥感影像森林类型深度学习精细分类方法:双支FCN-8s[J]. 林业科学, 2020, 56(3): 48-60.
[4]	赵霖,张晓丽,吴艳双,张斌. 面向机载高光谱数据的3D-CNN亚热带森林树种分类[J]. 林业科学, 2020, 56(11): 97-107.
[5]	张广群,李英杰,汪杭军,周厚奎. 基于微调CaffeNet的林业图像分类[J]. 林业科学, 2020, 56(10): 121-128.
[6]	陈龙现, 葛浙东, 罗瑞, 刘传泽, 刘晓平, 周玉成. 基于CNN的木材内部CT图像缺陷辨识[J]. 林业科学, 2018, 54(11): 127-133.