基于深度学习的木材优选锯视觉检测算法

doi:10.11707/j.1001-7488.20201214

摘要/Abstract

摘要：

目的: 提出一种基于深度学习的木材优选锯视觉检测算法，以最大限度增加木材出材率，提高木材价值，并进一步提升木材加工行业自动化水平。方法: 通过样本训练获得木材缺陷和木材等级识别网络，视觉传感器获取需检测木材图像，由木材缺陷识别网络确定木材缺陷的具体位置；对于无缺陷木材，由木材等级识别网络确定视觉传感器视场内木材的具体等级，进而确定切除部位在图像坐标系下的位置；由事先确定的图像平面与木材物理平面之间的单应关系确定木材最终切除位置列表。结果: 在本研究试验条件下，基于深度学习的木材优选锯视觉检测算法单幅图像缺陷检测时间为123 ms，缺陷检测正确率为95.8%，单幅图像等级分类识别时间为55 ms，分类识别正确率为97.1%，平均检测时间为86 ms，平均正确率为96.5%。结论: 基于深度学习的木材优选锯视觉检测算法运行速度快、识别准确率高、鲁棒性强，可克服传统优选锯分类不佳且需要人工干预的缺点，自动化程度高，能够满足木材优选锯实时、准确检测要求。

关键词: 深度学习, 优选锯, 木材处理, 缺陷识别, 计算机视觉

Abstract:

Objective: For the purposes of playing full value of wood, increasing the timber yield and improving the automatic level of wood processing industry, a new method for automatic optimizing cross-cut saw based on deep learning algorithm was proposed in this paper. Method: Pretrained defect detection network (Net.1) and grade classification network (Net.2) were obtained based on enough samples. Images of timber were captured by the vision sensor, the defect position on image could be confirmed based on Net. 1, while the grade position could be confirmed based on Net. 2. Final physical position could be determined based on the homography relationships between image plane and physical plane. Cutting list could also be obtained. Result: The experiment in our detailed conditions showed that the time of defect detection was 123 ms (per image), while the accuracy was 95.8% (per image). The time of grade classification was 55 ms (per image), while the accuracy was 97.1% (per image). The average time of detection was 86 ms (per image), while the average accuracy was 96.5% (per image). Conclusion: Our new method presented many advantages, such as a high operation speed, a high recognition efficiency and a strong robustness. The algorithm could improve the traditional optimizing cross-cut saw with a high automation and could also meet the requirements of real-time and accurate detection in wood processing.

Key words: deep learning, optimizing cross-cut saw, wood processing, defect detection, computer vision

中图分类号:

S777

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.

Mingwei Shao,Junyu Dong. A New Algorithm for Automatic Optimizing Cross-Cut Saw Based on Deep Learning Algorithm[J]. Scientia Silvae Sinicae, 2020, 56(12): 123-129.

图/表 10

图1

图2

图3

图4

图5

图6

图7

图8

图9

表1

参考文献 0

	马颂德, 张正友. 计算机视觉:计算理论与算法基础. 北京: 科学出版社, 1998.
	Ma S D , Zhang Z Y . Computer vision:computational theory and algorithm basis. Beijing: Science Press, 1998.
	Mitchell T M. 2012.机器学习.曾华军, 译.北京: 机械工业出版社.
	Mitchell T M. 2012. Machine learning. Translated by Zeng H J. Beijing: China Machine Press.[in Chinese]
	章毓晋. 图象工程:图象理解与计算机视觉. 北京: 清华大学出版社, 2000.
	Zhang Y J . Image engineering:image understanding and computer vision. Beijing: Tsinghua University Press, 2000.
	Dong C , Loy C C , He K M , et al. Image super-resolution using deep convolutional networks. IEEE Trans Pattern Anal Mach Intell, 2016, 38 (2): 295- 307. doi: 10.1109/TPAMI.2015.2439281
	Hartley R , Zisserman A . Multiple view geometry in computer vision. England: Cambridge University Press, 2003: 2- 8.
	He T , Liu Y , Xu C Y , et al. A fully convolutional neural network for wood defect location and identification. IEEE Access, 2019, doi: 10.1109/ACCESS.2019.2937461
	Huang H , Zhang H , Cheung Y . The common self-polar triangle of concentric circles and its application to camera calibration. IEEE Conference on Computer Vision and Pattern Recognition, 2015, 4065- 4072.
	Ismail H , Balachandran B . Algorithm fusion for feature extraction and map construction from SONAR data. IEEE Sensors Journal, 2015, 15 (11): 6460- 6471. doi: 10.1109/JSEN.2015.2456900
	Ji S W , Xu W , Yang M , et al. 3D convolutional neural networks for human action recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 35 (1): 221- 231.
	Razavian A S , Azizpour H , Carlsson S . CNN features off-the-shelf:an astounding baseline for recognition. IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2014, doi: 10.1109/CVPRW.2014.131
	Ren S Q , He K M , Girshick R , et al. Faster R-CNN:towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 39 (6): 1137- 1149.
	Safaee-Rad R , Tchoukanov I , Smith K C , et al. Three-dimensional location estimation of circular features for machine vision. IEEE Transactions on Robotics and Automation, 1992, 624- 640.
	Shao M W , Hu M J . Parallel feature based calibration method for a trinocular vision sensor. Optics Express, 2020a, 28 (14): 20573- 20586. doi: 10.1364/OE.393012
	Shao M W , Wang P , Wang J H . Improved sensors based on scheimpflug conditions and multi-focal constraints. IEEE Acess, 2020b, doi: 10.1109/ACCESS.2020.3020731
	Shao M W . Calibration methods for a camera with a tilted lens and a three-dimensional laser scanner in the scheimpflug condition. Journal of the Optical Society of America A:Optics, Image Science and Vision, 2020, 37 (7): 1076- 1082. doi: 10.1364/JOSAA.391906
	Shiu Y, Ahmad S. 1989.3D location of circular spherical features by monocular model-based vision. Proceedings of the IEEE Conference on System, Man and Cybernetics, Cambridge, Mass, USA, 576-581.
	Steger C. 1998. Unbiased extraction of curvilinear structures from 2D and 3D image. Muenchen: PhD Dissertation of Technische Universitaet Muenchen.
	Wu W , Gao X Y , Fan J , et al. Improved mask R-CNN-based cloud masking method for remote sensing images. International Journal of Remote Sensing, 2020, 41 (23): 8908- 8931.
	Zhang Z . A flexible new technique for camera calibration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22 (11): 1330- 1334. doi: 10.1109/34.888718
	Zheng W M . Heteroscedastic feature extraction for texture classification. IEEE Signal Processing Letters, 2009, 16 (9): 766- 769. doi: 10.1109/LSP.2009.2023939

缺陷检测Defect detection		等级分类Grade classification		平均指标Average result
单幅图像时间 Time per image/ms	正确率 Accuracy(%)	单幅图像时间 Time per image/ms	正确率 Accuracy(%)	平均时间 Average time/ms	平均正确率 Average accuracy(%)
123	95.8	55	97.1	86	96.5

[1]	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.
[2]	刘传泽, 王霄, 陈龙现, 郭慧, 罗瑞, 周玉成. 基于随机森林算法的纤维板表面缺陷识别[J]. 林业科学, 2018, 54(11): 121-126.
[3]	刘子豪, 祁亨年, 张广群, 汪杭军. 基于横切面微观构造图像的木材识别方法[J]. 林业科学, 2013, 49(11): 116-121.
[4]	汪杭军;汪碧辉. 一种新的针叶材自动识别方法[J]. 林业科学, 2011, 47(10): 141-145.
[5]	赵学增杨延竹王伟杰吴羡. 计算机技术在苗木自动分级中的应用发展概况[J]. 林业科学, 2004, 40(3): 162-166.