基于深度学习的75种阔叶材微观辨识方法

doi:10.11707/j.1001-7488.LYKX20240285

摘要/Abstract

摘要：

目的: 对75种南美进口阔叶材提出微观图像辨识模型TimberIDNet75，为海关、进出口检疫检验以及从事木材鉴定研究的人员提供一种准确的多材种辨识方法。方法: TimberIDNet75模型是包含1个输入层、4个隐含层的34层卷积层的中浅层神经网络。为尽量扩大感受野以提取更多图像特征，输入层采用13×13×256的卷积核，对每张图像提取256类特征，经激活、池化处理后作为输出。第1个隐含层采用2次卷积、激活后再进行残差修正，称作“两卷一修正块”。第1个隐含层包含3个“两卷一修正块”，提取256类特征作为输出。第2个隐含层包含4个“两卷一修正块”，再次提取512类特征作为输出。第3个隐含层包含6个“两卷一修正块”，对上一层的输出进行特征提取，获得1 024个类的特征。第4个隐含层包含3个“两卷一修正块”，对第3层的输出进行特征提取，获得2 048个类的特征，经全局平均池化后输入到全连接层映射出75个树种的分类。结果: TimberIDNet75模型的准确率达99.4%，损失值为0.044。将TimberIDNet75模型与现阶段较先进的深度学习模型进行比较，ResNet模型的准确率为98.1%、VGGNet模型的准确率为 97.1%、GoogleNet模型的准确率为96.2%、AlexNet模型的准确率为94.7%、ViT模型的准确率为53.2%，TimberIDNet75模型的准确率相比其中准确率最高的ResNet模型提高1.3%。利用TimberIDNet75模型对随机获取的75种进口阔叶材微观解剖样本进行实际测试，样本全部准确辨识，准确率达100%。结论: TimberIDNet75模型中的“两卷一修正块”，在节省机器资源的同时，可消除模型梯度下降导致过拟合的问题，同时利用残差法使得模型训练时人工干预降至最低，准确率和效率大幅提升。

关键词: 木材微观解剖, 深度学习网络模型, 木材材种辨识, 材种微观特征提取

Abstract:

Objective: A microscopic image recognition model TimberIDNet75 is proposed in this paper for 75 types of South American imported broadleaf timber, which provides an accurate method of multi-species recognition for customs, import and export quarantine inspection, and personnel engaged in timber identification research. Method: The TimberIDNet75 model is a medium-shallow neural network with 34 convolutional layers containing one input layer and four hidden layers. To expand the receptive field as much as possible to extract more image features, the input layer uses a 13×13×256 convolutional kernel to extract 256 types of features for each image, which are activated and pooled as the output. In the first hidden layer, two convolutions, activation and then residual correction are used, which is called“two-convolutions-one-correction-block”(TCOCB). The first hidden layer contains 3 TCOCBs, and 256 types of features are extracted as output. Then the second hidden layer contains 4 TCOCBs, extracting 512 classes of features as output. The third hidden layer contains 6 TCOCBs extracting features from the previous layer’s output, obtaining 1 024 classes of features. The fourth hidden layer contains 3 TCOCBs, after feature extraction from the output of layer 3, the features of 2 048 species were obtained, which were input into the fully connected layer after global average pooling to map the classification of 75 species. Result: The TimberIDNet75 model has an accuracy of 99.4% with a loss value of 0.044. Comparing the TimberIDNet75 model with the existing advanced deep learning models, ResNet model has an accuracy of 98.1%, VGGNet model has an accuracy of 97.1%, GoogleNet model has an accuracy of 96.2%, AlexNet model accuracy is 94.7%, ViT model accuracy is 53.2%, and TimberIDNet75 model accuracy is improved by 1.3% compared to the ResNet model, which has the highest accuracy among them. Then the TimberIDNet75 model was used to carry out practical tests on 75 randomly obtained microanatomical samples of imported broadleaf timber, and all the samples were accurately identified, with an accuracy rate of 100%. Conclusion: The TCOCB in the TimberIDNet75 model can eliminate the overfitting problem caused by model gradient drop while saving machine resources, meanwhile, the residual method makes it possible to minimize the manual intervention during model training, and the accuracy and efficiency are greatly improved.

Key words: wood microanatomy, deep learning network model, wood species identification, wood microanatomy feature extraction

中图分类号:

S781.1

田智康,葛浙东,郑焕祺,郑志帅,周玉成. 基于深度学习的75种阔叶材微观辨识方法[J]. 林业科学, 2024, 60(10): 94-103.

Zhikang Tian,Zhedong Ge,Huanqi Zheng,Zhishuai Zheng,Yucheng Zhou. Microscopic Identification Methods for 75 Types of Hardwood Based on Deep Neural Network[J]. Scientia Silvae Sinicae, 2024, 60(10): 94-103.

图/表 15

图1

表1

图2

图3

图4

图5

图6

图7

表2

表3

图8

表4

图9

表5

表6

参考文献 0

	程昱之, 钟丽辉, 孙永科. 木材识别技术研究综述. 农业技术与装备, 2021, (1): 125- 128. doi: 10.3969/j.issn.1673-887X.2021.01.054
	Cheng Y Z, Zhong L H, Sun Y K. Review of wood identification technology. Agricultural Technology & Equipment, 2021, (1): 125- 128. doi: 10.3969/j.issn.1673-887X.2021.01.054
	程昱之, 钟丽辉, 何　鑫, 等. 基于改进K-means聚类与水平集的木材横截面管孔分割. 森林工程, 2022, 38 (1): 42- 51.
	Cheng Y Z, Zhong L H, He X, et al. Segmentation of wood cross-section pore based on improved K-means clustering and level-set. Forest Engineering, 2022, 38 (1): 42- 51.
	戴天虹, 翟　冰. 基于改进EfficientNet的木材识别研究. 森林工程, 2023, 39 (4): 93- 100.
	Dai T H, Zhai B. Wood recognition research based on improved EfficientNet. Forest Engineering, 2023, 39 (4): 93- 100.
	付立岩, 冯国红, 刘旭铭. 基于多元经验模态分解的可见/近红外光谱识别木材研究. 森林工程, 2023, 39 (4): 101- 109.
	Fu L Y, Feng G H, Liu X M. Wood recognition by visible/near infrared spectroscopy based on multivariate empirical mode decomposition. Forest Engineering, 2023, 39 (4): 101- 109.
	李　楠. 2018. 一种基于卷积神经网络的轻量级木材图像识别模型研究. 杭州: 浙江农林大学.
	Li N. 2018. A lightweight wood image recognition model based on convolutional neural networks. Hangzhou: Zhejiang A & F University［in Chinese］
	宿恒硕, 吕　军, 丁志平, 等. 基于改进残差神经网络的木材识别算法. 林业科学, 2021, 57 (12): 147- 154.
	Su H S, Lü J, Ding Z P, et al. Wood identification algorithm based on improved residual neural network. Scientia Silvae Sinicae, 2021, 57 (12): 147- 154.
	杨霄霞. 2022. 基于红木微观结构的分类与识别算法研究. 济南: 山东建筑大学.
	Yang X X. 2022. Research on classification and recognition algorithm based on rosewood microstructure. Jinan: Shandong Jianzhu University. ［in Chinese］
	尹江苹, 蒋劲东, 高瑞清, 等. CITES公约木材树种管制及我国进口濒危木材贸易现状. 木材工业, 2019, 33 (1): 25- 28, 37.
	Yin J P, Jiang J D, Gao R Q, et al. Current status of CITES listed timber species and relevant imports to China. China Wood Industry, 2019, 33 (1): 25- 28, 37.
	赵子宇, 杨霄霞, 郭　慧, 等. 基于卷积神经网络模型的木材宏、微观辨识方法. 林业科学, 2021, 57 (6): 134- 143. doi: 10.11707/j.1001-7488.20210615
	Zhao Z Y, Yang X X, Guo H, et al. Recognition method of wood macro-and micro-structure based on convolution neural network. Scientia Silvae Sinicae, 2021, 57 (6): 134- 143. doi: 10.11707/j.1001-7488.20210615
	Filho P L P, Oliveira L S, Nisgoski S, et al. Forest species recognition using macroscopic images. Machine Vision and Applications, 2014, 25 (4): 1019- 1031. doi: 10.1007/s00138-014-0592-7
	Geirhos R, Jacobsen J H, Michaelis C, et al. Shortcut learning in deep neural networks. Nature Machine Intelligence, 2020, 2, 665- 673. doi: 10.1038/s42256-020-00257-z
	Haralick R M, Shanmugam K, Dinstein I. 1973. Textural features for image classification. IEEE Transactions on Systems, Man, and Cybernetics, SMC-3(6): 610-621.
	He K M, Zhang X Y, Ren S Q, et al. 2016. Deep residual learning for image recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV: USA, IEEE, 770−77.
	Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks. International Conference on Neural Information Processing Systems, 2012, 25 (2): 1097- 1105.
	LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 1998, 86 (11): 2278- 2324. doi: 10.1109/5.726791
	Lens F, Liang C, Guo Y H, et al. Computer-assisted timber identification based on features extracted from microscopic wood sections. IAWA Journal, 2020, 41 (4): 660- 680. doi: 10.1163/22941932-bja10029
	Martins J, Oliveira L S, Nisgoski S, et al. A database for automatic classification of forest species. Machine Vision and Applications, 2013, 24 (3): 567- 578. doi: 10.1007/s00138-012-0417-5
	Moulin J C, Lopes D J V, Mulin L B, et al. Microscopic identification of Brazilian commercial wood species via machine-learning. CERNE, 2022, 28 (1): 1- 9.
	Ojala T, Pietikäinen M, Harwood D. A comparative study of texture measures with classification based on featured distributions. Pattern Recognition, 1996, 29 (1): 51- 59. doi: 10.1016/0031-3203(95)00067-4
	Olshausen B A, Field D J. Emergence of simple-cell receptive field properties by learning a sparse code for natural images. Nature, 1996, 381 (6583): 607- 609. doi: 10.1038/381607a0
	Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research, 2014, 15 (1): 1929- 1958.

树种名称 Species name	裁切图像 Cropped image	旋转90° Rotate 90°	旋转180° Rotate 180°	旋转270° Rotate 270°	垂直翻转 Flip vertical
金叶山榄 Chrysophyllum sp.
白花翅玉蕊 Cariniana estrellensis
纤皮玉蕊 Couratari sp.
银桦 Grevillea robusta

实际值Actual value	预测值Predicted value
实际值Actual value	预测为真True	预测为假Negative
实际为真True	TP（11 185）	FN（65）
实际为假False	FP（0*）	TN（0*）

预测树种名称 Predicted species name	预测树种编号 Predicted species ID	实际树种名称 True species label	实际树种编号 True species ID	分类错误图像张数 Misclassified pictures
纤皮玉蕊Couratari sp.	2	大叶桃花心木Swietenia macrophylla	70	1
纸叶猴钵玉蕊Eschweleira chartaceae	4	孪叶苏木Hymenaea courbaril	10	14
镰荚豆木Eperua falcata	9	孪叶苏木Hymenaea courbaril	10	2
镰荚豆木 Eperua falcata	9	红卡雅楝木Khaya ivorensis	68	1
堇色紫檀Pterocarpus violaceus	13	镰荚豆木Eperua falcata	9	1
蛇状柯那豆木Anadenanthera colubrina	15	梯叶香脂树Copaifera trapezifolia	8	1
亚马孙黄檀Dalbergia spruceana	18	革质拉美玉蕊木Eschweilera matamata	3	1
亚马孙黄檀Dalbergia spruceana	18	孪叶苏木Hymenaea courbaril	10	1
异味豆木Dinizia excelsa	20	肥果桃榄Pouteria pachycarpa	7	1
银合欢Leucaena leucocephala	23	金叶山榄Chrysophyllum sp.	5	1
小叶含羞树Mimosa scabrella	26	苞序楠Virola oleifera	45	4
坚硬赛落腺豆木Parapiptadenia rigida	28	苞序楠Virola oleifera	45	4
大落腺豆木Piptadenia excelsa	30	梯叶香脂树Copaifera trapezifolia	8	1
重蚁木Tabebuia sp.	35	大果泡泽木Porcelia macrocarpa	41	6
巨桉Eucalyptus grandis	48	大果柯那豆木Anadenanthera peregrina	16	1
巨桉Eucalyptus grandis	48	因加豆树Inga sessilis	22	14
巨桉Eucalyptus grandis	48	柳叶桉Eucalyptus saligna	49	3
轴状独蕊木Erisma uncinatum	51	夸雷木Qualea sp.	52	1
北枳椇Hovenia dulcis	60	苞序楠Virola oleifera	45	2
欧鼠李 Rhamnus frangula	61	光荚含羞树Mimosa bimucronata	25	5

树种名称Species name	辨识输出Model output	重复张数Test number	准确率Accuracy(%)	识别时间Time/s
金叶山榄	043 Chrysophyllum sp.	3	100	0.70
白花翅玉蕊	039 Cariniana estrellensis	1	100	0.22
白花翅玉蕊	039 Cariniana estrellensis	2	100	0.30
大果柯那豆木	054 Anadenanthera peregrina	3	100	0.71
银桦	092 Grevillea robusta	1	100	0.21
银桦	092 Grevillea robusta	2	100	0.46
银桦	092 Grevillea robusta	3	100	0.69
苦樗	112 Simaruba amara	3	100	0.79
纤皮玉蕊	040 Couratari sp.	3	100	0.61
堇色紫檀	051 Pterocarpus violaceus	3	100	0.72
白花翅玉蕊	039 Cariniana estrellensis	1	100	0.23

[1]	苗宇杰,祝诗平,普京,李俊贤,马羚凯,黄华. 基于卷积神经网络的家具木材图像种类识别[J]. 林业科学, 2023, 59(8): 133-140.
[2]	杨楠楠,白岩,姜苏译,杨春梅,徐凯宏. 基于AlexNet优化的板材材种识别方法[J]. 林业科学, 2022, 58(3): 149-158.
[3]	宿恒硕,吕军,丁志平,唐彦杰,陈旭东,周强,张哲宇,姚青. 基于改进残差神经网络的木材识别算法[J]. 林业科学, 2021, 57(12): 147-154.
[4]	袁霄,陈勇平,唐启恒,郭文静. 基于微损探测的古建筑木构件常用树种识别[J]. 林业科学, 2021, 57(12): 122-131.
[5]	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.
[6]	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.
[7]	刘衡,施福军,黎韦水,韦中绵,马若克,刘志高,李英健,符韵林. 降香黄檀与东京黄檀木材解剖特征及抽提物化学成分的可区别性[J]. 林业科学, 2021, 57(2): 103-114.
[8]	韩彪,李文清,郭素娟,陆璐,解孝满. 基于差示扫描量热技术的板栗胚轴低温保存技术及临界含水量[J]. 林业科学, 2020, 56(3): 21-27.
[9]	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.
[10]	葛浙东, 戚玉涵, 罗瑞, 陈龙现, 王艳伟, 周玉成. 木材CT断层成像系统旋转中心校正方法[J]. 林业科学, 2018, 54(11): 164-171.
[11]	秦理哲, 林兰英, 傅峰. 木材胶合界面微观结构样品制备新方法——激光烧蚀技术[J]. 林业科学, 2018, 54(4): 93-99.
[12]	郭素娟;谢鹏. 外源物质对板栗淀粉合成及相关酶活性的影响[J]. 林业科学, 2013, 49(4): 135-140.
[13]	张广群;汪杭军. 基于多目标遗传算法的管孔组合特征识别[J]. 林业科学, 2012, 48(4): 87-92.
[14]	赵天田;梁丽松;林顺顺;张柏林;王贵禧. 板栗淀粉表面结合脂中脂肪酸的组成[J]. 林业科学, 2012, 48(2): 96-102.
[15]	杨淑敏;刘杏娥;费本华;江泽慧;杨喜;单海斌. 茶秆竹细胞壁解剖特性以及木质素微区分布[J]. 林业科学, 2012, 48(2): 129-133.