基于深度学习的结构用锯材节子在线检测方法

doi:10.11707/j.1001-7488.LYKX20250012

摘要/Abstract

摘要：

目的: 针对结构用锯材人工目测分等效率低、主观性强等问题，选取常用赤松结构用锯材，构建一种基于深度学习的节子在线检测方法，为提高结构用锯材分等的自动化与精度提供技术支撑。方法: 基于 YOLO 网络构建节子检测模型，引入高效层聚合网络（ELAN）以及结合 SIFT 特征的图像拼接、分割与融合方法，增强机器视觉缺陷检测系统在锯材分等及复杂生产线视觉任务中的适应能力。通过多尺度预测和损失函数最小化，抑制图像背景及噪声对缺陷检测的干扰，准确计算目标分类与定位损失，从而提高节子目标检测精度并优化模型在特定任务中的表现。结果: 工业现场应用结果表明，该系统对锯材表面节子缺陷的识别与检测精度达到90.97%，漏检率为9.03%，节子位置（X，Y）与尺寸（L，W）平均检测精度分别为86.29%与85.95%，检测速度可达到 20~30 m·min^?1，能够满足结构用锯材在加工生产线中的应用需求。结论: 深度学习方法适用于实际锯材检测任务，可有效降低人工检测的主观性，并提升检测的准确率和效率。机器视觉检测技术的引入，有望推动木材分等技术的创新与发展，提升木材加工行业的质量控制水平，进而促进木结构建筑行业整体技术能力的进步。

关键词: 赤松结构用锯材, 机器视觉节子检测平台, 深度学习, YOLO 训练模型, 高效层聚合网络

Abstract:

Objective: To address the low efficiency and strong subjectivity of manual visual grading for structural sawn timber, this study selected Pinus densiflora structural sawn timber and developed a deep-learning-based online knot detection method, providing technical support for improving the automation and accuracy of structural sawn timber grading. Method: Based on the YOLO network, a knot-detection model was built that incorporates the efficient layer aggregation network (ELAN) and an image stitching, segmentation, and fusion scheme guided by SIFT features, strengthening the adaptability of the machine-vision defect-detection system to sawn-timber grading and other complex on-line vision tasks. Multi-scale prediction and loss-function minimization suppress background clutter and noise, enabling accurate classification and localization losses and thus raising knot-detection accuracy while optimizing task-specific performance. Result: In the industrial production application site, the test results showed that the identification and defect detection accuracy of knots on the surface of lumber is 90.97%, with a missed detection rate of 9.03%. The average detection accuracy of knot defect location X and Y was 86.29%, the average detection accuracy of knot defect size L and W was 85.95%, and the detection speed could reach 20–30 m·min^?1, which meets the practical application of wooden product processing line. Conclusion: In this study, the deep learning method is suitable for lumber detection in practical application, reducing the subjectivity of manual inspection and improving the accuracy and efficiency of inspection. Machine vision detection technique promotes the innovation and development of wood grading technology, improves the quality of wood processing industry, and improves the technical level of wood structure construction industry.

Key words: Pinus densiflora structural lumber, machine vision knot inspection platform, deep learning, YOLO training model, efficient layer aggregation network

中图分类号:

S777

纪敏,高锐,王晓欢,刁兴良,韩佳锴,赵扬,王国富,张伟. 基于深度学习的结构用锯材节子在线检测方法[J]. 林业科学, 2025, 61(11): 150-159.

Min Ji,Rui Gao,Xiaohuan Wang,Xingliang Diao,Jiakai Han,Yang Zhao,Guofu Wang,Wei Zhang. Online Detecting Method of Structural Lumber Knot Based on Deep Learning[J]. Scientia Silvae Sinicae, 2025, 61(11): 150-159.

图/表 12

图1

图2

图3

图4

图5

图6

图7

图8

表1

图9

表2

表3

参考文献 0

	陈龙现, 刘传泽, 杜光月, 等. 人造板在线同步图像采集系统设计与实现. 制造业自动化, 2019, 41 (1): 68- 71. doi: 10.3969/j.issn.1009-0134.2019.01.016
	Chen L X, Liu C Z, Du G Y, et al. Design and implementation of synchronous image acquisition system of wood-based panels. Manufacturing Automation, 2019, 41 (1): 68- 71. doi: 10.3969/j.issn.1009-0134.2019.01.016
	邓斌攸,池志强,潘云峰,等. 家具板件圆形孔位的机器视觉在线检测算法. 木材科学与技术, 2022, 36 (2): 60- 64. doi: 10.12326/j.2096-9694.2021099
	Deng B Y, Chi Z Q, Pan Y F, et al. Algorithm of machine vision system for detecting holes in furniture panel. Chinese Journal of Wood Science and Technolog, 2022, 36 (2): 60- 64. doi: 10.12326/j.2096-9694.2021099
	侯晓鹏, 吴智慧, 刘海波, 等. 基于OPC统一架构的木制品加工设备信息互联互通方法. 木材科学与技术, 2022, 36 (6): 95- 102. doi: 10.12326/j.2096-9694.2022098
	Hou X P, Wu Z H, Liu H B, et al. Information interconnection and intercommunication method for woodworking equipment based on open platform communications unified architecture(OPC UA). Chinese Journal of Wood Science and Technology, 2022, 36 (6): 95- 102. doi: 10.12326/j.2096-9694.2022098
	贾浩男, 徐华东, 王立海, 等. 基于改进YOLOv5木板材表面缺陷的定量识别. 北京林业大学学报, 2023, 45 (4): 147- 155. doi: 10.12171/j.1000-1522.20220419
	Jia H N, Xu H D, Wang L H, et al. Quantitative identification of surface defects in wood paneling based on improved YOLOv5. Journal of Beijing Forestry University, 2023, 45 (4): 147- 155. doi: 10.12171/j.1000-1522.20220419
	彭晓瑞, 吕　斌, 李伟光, 等. 我国板式家具先进制造技术研发与应用进展. 木材科学与技术, 2023, 37 (2): 1- 7, 15. doi: 10.12326/j.2096-9694.2022085
	Peng X R, Lyu B, Li W G, et al. Development and application progress of advanced manufacturing technology for China’s panel furniture. Chinese Journal of Wood Science and Technology, 2023, 37 (2): 1- 7, 15. doi: 10.12326/j.2096-9694.2022085
	田智康, 葛浙东, 郑焕祺, 等. 2024. 基于深度学习的75种阔叶材微观辨识方法. 林业科学, 60(10): 94−103.
	Tian Z K, Ge Z D, Zheng H Q, et al. 2024. Microscopic identification methods for 75 types of hardwood based on deep neural network. Scientia Silvae Sinicae, 60(10): 94−103. ［in Chinese］
	王　勇, 张　伟. 锯材表面缺陷识别方法对比分析研究. 世界林业研究, 2022, 35 (4): 47- 52.
	Wang Y, Zhang W. Comparative analysis of surface defect recognition methods for sawntimber. World Forestry Research, 2022, 35 (4): 47- 52.
	王　晰, 简振雄, 任明俊. 基于深度学习的逆向反射模型. 光学学报, 2023, 43 (21) doi: 10.3788/AOS230615
	Wang X, Jian Z X, Ren M J. Inverse reflectance model based on deep learning. Acta Optica Sinica, 2023, 43 (21) doi: 10.3788/AOS230615
	王晓明, 邓　璐, 史一哲, 等. 基于Harris特征与NDT-ICP算法的钢箱拱预制件尺寸智检方法. 交通运输工程学报, 2024, 24 (1): 158- 170.
	Wang X M, Deng L, Shi Y Z, et al. Intelligent dimensional inspection method for steel box arch prefabricated components based on Harris features and NDT-ICP algorithm. Journal of Traffic and Transportation Engineering, 2024, 24 (1): 158- 170.
	吴雨生, 张　伟, 候晓鹏, 等. 基于机器视觉的杉木规格材节子检测与分等. 林业和草原机械, 2020, 1 (2): 13- 16.
	Wu Y S, Zhang W, Hou X P, et al. Fir structural timber detecting and grading based on machine vision technology. Forestry and Grassland Machinery, 2020, 1 (2): 13- 16.
	杨　凡, 杨博凯, 李荣荣. 基于图像分割和深度学习的人造板表面缺陷检测. 浙江农林大学学报, 2024, 41 (1): 176- 182.
	Yang F, Yang B K, Li R R. Surface defect detection technology of wood-based panel based on image segmentation and deep learning. Journal of Zhejiang A & F University, 2024, 41 (1): 176- 182.
	喻　炜, 周海燕, 刘　英, 等. 人造板无损检测技术研究进展. 世界林业研究, 2023, 36 (3): 58- 62.
	Yu W, Zhou H Y, Liu Y, et al. Research progress in nondestructive testing technology for wood-based panels. World Forestry Research, 2023, 36 (3): 58- 62.
	朱剑刚, 王　旭. 木质家具智能制造赋能技术及发展路径分析. 林业工程学报, 2021, 6 (6): 177- 183.
	Zhu J G, Wang X. Research on enabling technologies and development path of intelligent manufacturing of wooden furniture. Journal of Forestry Engineering, 2021, 6 (6): 177- 183.
	张　赛, 王应彪, 杨　谭, 等. 基于改进LeNet-5模型的木材表面典型缺陷识别方法研究. 木材科学与技术, 2021, 35 (6): 31- 37. doi: 10.12326/j.2096-9694.2021024
	Zhang S, Wang Y B, Yang T, et al. Detecting method of the wood surface defects based on modified LeNet-5 model. Chinese Journal of Wood Science and Technology, 2021, 35 (6): 31- 37. doi: 10.12326/j.2096-9694.2021024
	Barmpoutis P, Dimitropoulos K, Barboutis I, et al. Wood species recognition through multidimensional texture analysis. Computers and Electronics in Agriculture, 2018, 144, 241- 248. doi: 10.1016/j.compag.2017.12.011
	Bengio Y, LeCun Y, Hinton G. Deep learning for AI. Communications of the ACM, 2021, 64 (7): 58- 65. doi: 10.1145/3448250
	Brunetti M, Burato P, Cremonini C, et al. Visual and machine grading of larch (Larix decidua Mill. ) structural timber from the Italian Alps. Materials and Structures, 2016, 49 (7): 2681- 2688. doi: 10.1617/s11527-015-0676-5
	Chung S, Abbott L F. Neural population geometry: an approach for understanding biological and artificial neural networks. Current Opinion in Neurobiology, 2021, 70, 137- 144. doi: 10.1016/j.conb.2021.10.010
	Gazo R, Wells L, Krs V, et al. Validation of automated hardwood lumber grading system. Computers and Electronics in Agriculture, 2018, 155, 496- 500. doi: 10.1016/j.compag.2018.06.041
	Ishida T, Yamane I, Sakai T, et al. 2020. Do we need zero training loss after achieving zero training error? arXiv preprint arXiv: 2002.08709.
	Ji M, Zhang W, Wang G F, et al. Machine vision for knot detection and location in Chinese fir lumber. Forest Products Journal, 2024, 74 (2): 185- 202. doi: 10.13073/FPJ-D-23-00050
	Ji M, Zhang W, Han J K, et al. A deep learning-based algorithm for online detection of small target defects in large-size sawn timber. Industrial Crops and Products, 2024, 222, 119671. doi: 10.1016/j.indcrop.2024.119671
	Kamal K, Qayyum R, Mathavan S, et al. Wood defects classification using laws texture energy measures and supervised learning approach. Advanced Engineering Informatics, 2017, 34, 125- 135. doi: 10.1016/j.aei.2017.09.007
	Li X Y, Chiong R, Hu Z Y, et al. A graph neural network model with local environment pooling for predicting adsorption energies. Computational and Theoretical Chemistry, 2023, 1226, 114161. doi: 10.1016/j.comptc.2023.114161
	Lukacevic M, Kandler G, Hu M, et al. A 3D model for knots and related fiber deviations in sawn timber for prediction of mechanical properties of boards. Materials & Design, 2019, 166, 107617.
	Pahlberg T, Hagman O, Thurley M. Recognition of boards using wood fingerprints based on a fusion of feature detection methods. Computers and Electronics in Agriculture, 2015, 111, 164- 173. doi: 10.1016/j.compag.2014.12.014
	Pang G S, Shen C H, Cao L B, et al. Deep learning for anomaly detection. ACM Computing Surveys, 2022, 54 (2): 1- 38.
	Ramos L T, Casas E, Romero C, et al. A study of YOLO architectures for wildfire and smoke detection in ground and aerial imagery. Results in Engineering, 2025, 26, 104869. doi: 10.1016/j.rineng.2025.104869
	Tu Y X, Ling Z G, Guo S Y, et al. An accurate and real-time surface defects detection method for sawn lumber. IEEE Transactions on Instrumentation and Measurement, 2020, 70, 2501911.
	Wang B. 2022. A parallel implementation of computing mean average precision. arXiv preprint arXiv, 2206.09504.
	Wang Y, Zhang W, Gao R, et al. Recent advances in the application of deep learning methods to forestry. Wood Science and Technology, 2021, 55 (5): 1171- 1202. doi: 10.1007/s00226-021-01309-2
	Wright L G, Onodera T, Stein M M, et al. Deep physical neural networks trained with backpropagation. Nature, 2022, 601, 549- 555. doi: 10.1038/s41586-021-04223-6
	Zhao J B, Meng Z X, Jin Z, et al. Bending properties of bamboo scrimber with holes in different sizes and positions. Construction and Building Materials, 2019, 200, 209- 217. doi: 10.1016/j.conbuildmat.2018.12.076

样品分组 Sample grouping	2个宽面 Two wide face	实际节子数量 The actual number of knots / piece	检出节子数量 Number of detected knots / piece	漏检率 Missed detection rate（%）	非节子但被识别为节子数量 Misjudged as a knot/ piece	检测精度 Detection precision（%）
A	A	12	11	8.33	1	91.67
A	B	14	14	0.00	0	100.00
B	A	5	4	20.00	0	80.00
B	B	4	4	0.00	0	100.00
C	A	15	13	13.33	0	86.67
C	B	16	13	18.75	1	81.25
D	A	8	7	12.50	1	87.50
D	B	7	6	14.29	0	85.71
E	A	13	12	7.69	2	92.31
E	B	14	12	14.29	1	85.71
F	A	5	5	0.00	0	100.00
F	B	7	6	14.29	1	85.71
G	A	10	10	0.00	0	100.00
G	B	13	11	15.38	0	84.62
H	A	4	4	0.00	0	100.00
H	B	6	6	0.00	0	100.00
I	A	15	13	13.33	1	86.67
I	B	12	11	8.33	1	91.67
J	A	5	4	20.00	0	80.00
J	B	4	4	0.00	1	100.00

检测指标 Detection metrics	精度Precision（%）
识别与检测精度 Detection accuracy	90.97
漏检率 Missed detection rate	9.03
模型均值平均精度 Model mAP	91.50
节子缺陷位置（X，Y）平均检测精度 Average detection accuracy of defect location （X， Y）	86.29
节子缺陷尺寸（L，W）平均检测精度 Average detection accuracy of defect size （L， W）	85.95

方法 Method	均值平均精度mAP（%）	识别与检测精度 Detection accuracy（%）	测量耗时 Measurement time/s
图像预处理与特征提取 Image preprocessing and feature extraction	—	75.12	2.10
深度学习方法 Deep learning methods	95	90.97	2.00
人工测量 Manual measurement	—	91.56	22.80

[1]	葛晓宁,许新桥,张怀清,张京,杨杰,崔泽宇,傅汝饶,梁金洁,邹添华,王林龙,刘洋. 林木基因型-环境互作算法研究进展与思考[J]. 林业科学, 2025, 61(3): 1-15.
[2]	谢婉滢,刘文萍,王晗. 无人机松林图像早期松材线虫病害检测[J]. 林业科学, 2024, 60(9): 124-133.
[3]	齐建东,郑尚姿,陈子仪,马鐘添. 基于ConvNeXt的北京地区红外相机野生动物图像识别改进模型构建[J]. 林业科学, 2024, 60(8): 33-45.
[4]	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.
[5]	齐建东,马鐘添,张德怀,田赟. 基于BS-ResNeXt-50的密云地区野生动物图像识别[J]. 林业科学, 2023, 59(8): 112-122.
[6]	李贾,兰岚,张佐忠,袁文涛,高德民,宗树琴,业巧林. 基于深度学习的森林可燃物含水率反演技术[J]. 林业科学, 2022, 58(10): 47-58.
[7]	周焱,刘文萍,骆有庆,宗世祥. 基于深度学习的小目标受灾树木检测方法[J]. 林业科学, 2021, 57(3): 98-107.
[8]	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.
[9]	邵明伟,董军宇. 基于深度学习的木材优选锯视觉检测算法[J]. 林业科学, 2020, 56(12): 123-129.