欢迎访问作物学报,今天是

作物学报 ›› 2020, Vol. 46 ›› Issue (11): 1771-1779.doi: 10.3724/SP.J.1006.2020.94187

• 耕作栽培·生理生化 • 上一篇    下一篇

基于深度学习的大豆豆荚类别识别研究

闫壮壮2(), 闫学慧2, 石嘉2, 孙凯2, 虞江林2, 张战国1, 胡振邦3, 蒋鸿蔚3, 辛大伟3, 李杨1, 齐照明3, 刘春燕3, 武小霞3, 陈庆山3, 朱荣胜1,*()   

  1. 1 东北农业大学文理学院, 黑龙江哈尔滨 150030
    2 东北农业大学工程学院, 黑龙江哈尔滨 150030
    3 东北农业大学大豆研究所, 黑龙江哈尔滨 150030
  • 收稿日期:2019-11-26 接受日期:2020-07-02 出版日期:2020-11-12 网络出版日期:2020-07-13
  • 通讯作者: 朱荣胜
  • 作者简介:E-mail:zhuangyanneau@163.com
  • 基金资助:
    本研究由国家自然科学基金项目(31471516);国家自然科学基金青年项目(31400074)

Classification of soybean pods using deep learning

YAN Zhuang-Zhuang2(), YAN Xue-Hui2, SHI Jia2, SUN Kai2, YU Jiang-Lin2, ZHANG Zhan-Guo1, HU Zhen-Bang3, JIANG Hong-Wei3, XIN Da-Wei3, LI Yang1, QI Zhao-Ming3, LIU Chun-Yan3, WU Xiao-Xia3, CHEN Qing-Shan3, ZHU Rong-Sheng1,*()   

  1. 1 College of Arts and Sciences, Northeast Agricultural University, Harbin 150030, Heilongjiang, China
    2 Engineering College Northeast Agricultural University, Harbin 150030, Heilongjiang, China
    3 Soybean Research Institute, Northeast Agricultural University, Harbin 150030, Heilongjiang, China
  • Received:2019-11-26 Accepted:2020-07-02 Published:2020-11-12 Published online:2020-07-13
  • Contact: Rong-Sheng ZHU
  • Supported by:
    This study was supported by the National Natural Science Foundation of China(31471516);the National Natural Science Foundation of China Youth Project(31400074)

摘要:

作物表型调查是作物品种选育过程中的一项关键工作。传统表型调查主要依靠人力,使得表型调查的结果难以达到自动化、高精度、高可靠性的要求。在大豆的表型调查中,对豆荚类别的正确识别是豆荚个数、长度和宽度等表型准确提取的关键和前提。本文针对成熟期大豆豆荚的图片, 通过利用深度学习迁移5种不同的网络模型[AlexNet、VggNet (Vgg16, Vgg19)、GoogleNet、ResNet-50], 对一粒荚、二粒荚、三粒荚、四粒荚进行识别。为提高训练速度和准确率, 本试验微调模型, 选择不同的优化器(SGD、Adam)对网络模型进行优化。结果表明, 在针对豆荚辨识问题中, Adam的性能优于SGD, 而Vgg16网络模型搭配Adam优化器, 豆荚类别的测试准确率达到了98.41%, 在所选的网络模型中体现了最佳的性能。在十折交叉验证试验中也体现了Vgg16网络模型具有良好的稳定性。因此本研究认为Vgg16网络模型可以应用到实际的豆荚识别中, 为进一步实现豆荚表型自动提取提供一条重要的解决途径。

关键词: 大豆育种, 豆荚辨别, 深度学习, 迁移学习

Abstract:

Crop phenotype investigation is a key task in the selection and breeding of crop varieties. The traditional phenotypic survey mainly relies on human labors, which makes the results of the phenotypic survey difficult to meet the requirements of automation, high precision and high reliability. In the investigation of soybean phenotypes, the correct identification of pod types is the key and premise for the accurate extraction of phenotypes such as the number, length and width of pods. This study focused on the pictures of mature soybean pods by using deep learning to migrate five different network models [AlexNet, VggNet (Vgg16, Vgg19), GoogleNet, ResNet-50], to identify one-pod, two-pod, three-pod, and four-pod. In order to improve training speed and accuracy, this experiment fine-tuning the model and selected different optimizers (SGD, Adam) to optimize the network model. Adam’s performance was better than SGD in the problem of pod identification. With the Vgg16 network model and the Adam optimizer, the test accuracy of the pod category reached 98.41%, which reflected the best performance in the selected network model. In the 10-fold cross-validation test, the Vgg16 network model had good stability. Therefore, this study indicates that the Vgg16 network model can be applied to the actual identification of pods, and provide an important solution for further automatic extraction of pod phenotypes.

Key words: soybean breeding, pod identification, deep learning, transfer learning

图1

数据增强 a: 左右翻转时针旋转90度; b: 逆时针旋转90度; c: 调亮度; d: 左右翻转; e: 上下翻转; f: 调整饱和度; g: 逆时针旋转180度。"

图2

AlexNet网络结构"

图3

深度网络模型在不同算法下的表现 a: AlexNet模型在Adam算法和SGD算法下的验证准确率; b: Vgg16模型在Adam算法和SGD算法下的验证准确率; c: GoogleNet模型在Adam算法和SGD算法下的验证精准确率; d: ResNet-50模型在Adam算法和SGD算法下的验证准确率。横轴为迭代次数, 纵轴为验证准确率。"

图4

模型准确率和损失图 a: 5种模型搭配Adam训练算法在相同数据集下的验证准确率; b: 5种模型搭配Adam训练算法在相同数据集下的验证损失。"

表1

不同迁移模型搭配Adam的试验结果"

模型
Model
层数
Layer
迭代次数
Iteration
时间
Time (min)
准确率
Accuracy (%)
AlexNet (Adam) 8 2000 22.48 97.11
Vgg16 (Adam) 16 2000 144.56 98.41
Vgg19 (Adam) 19 2000 255.23 98.35
GoogleNet (Adam) 22 2000 18.13 95.83
ResNet-50 (Adam) 50 2000 24.25 87.15

表2

十折交叉验证结果"

折次
k-fold
验证准确率
Accuracy
平均损失
Average loss
1-fold 0.9652 0.1574
2-fold 0.9552 0.2024
3-fold 0.9788 0.1779
4-fold 0.9623 0.1307
5-fold 0.9811 0.1523
6-fold 0.9623 0.2104
7-fold 0.9735 0.1903
8-fold 0.9782 0.1531
9-fold 0.9717 0.1658
10-fold 0.9693 0.1675
均值Mean 0.9697 0.1707
标准差Standard deviation 0.0085 0.0234

图5

模型混淆矩阵热力图 混淆矩阵中列为预测类别, 行为真实类别。图最右边的列为精确率, 底部的行为召回率。"

表3

模型的分类性能"

豆荚种类
Pod class
精确率
Precision (%)
召回率
Recall (%)
F1分数
F1-score (%)
一粒荚One-pod 99.5 98.1 98.8
二粒荚Two-pod 95.8 96.8 96.3
三粒荚Three-pod 95.6 95.6 95.6
四粒荚Four-pod 98.3 98.6 98.4
[1] Tester M, Langridge P. Breeding technologies to increase crop production in a changing world. Science, 2010,327:818-822.
pmid: 20150489
[2] Ribaut J M, Vicente M D, Delannay X. Molecular breeding in developing countries: challenges and perspectives. Curr Opin Plant Biol, 2010,13:213-218.
pmid: 20106715
[3] Furbank R T, Tester M. Phenomics-technologies to relieve the phenotyping bottleneck. Trends Plant Sci, 2011,16:635-644.
pmid: 22074787
[4] Cannon S B, Jackson M S A. Three sequenced legume genomes and many crop species: rich opportunities for translational genomics. Plant Physiol, 2009,151:970-977.
pmid: 19759344
[5] Rahaman M M, Chen D J, Gillani Z, Klukas C, Chen M. Advanced phenotyping and phenotype data analysis for the study of plant growth and development. Front Plant Sci, 2015,6:619.
pmid: 26322060
[6] Shakoor N, Lee S, Mockler T C. High throughput phenotyping to accelerate crop breeding and monitoring of diseases in the field. Curr Opin Plant Biol, 2017,38:184-192.
pmid: 28738313
[7] 邱丽娟. 大豆种质资源描述规范和数据标准. 北京: 中国农业出版社, 2006. pp 64-74.
Qiu L J. Specification and Data Standard of Soybean Germplasm Resources. Beijing: China Agriculture Press, 2006. pp 64-74(in Chinese).
[8] Lecun Y, Bengio Y, Hinton G. Deep learning. Nature, 2015,521:436.
pmid: 26017442
[9] Mohanty S P, Hughes D P, Salathé M. Using deep learning for image-based plant disease detection. Front Plant Sci, 2016,7:1419.
[10] Cheng X, Zhang Y, Chen Y, Wu Y, Yue Y. Pest identification via deep residual learning in complex background. Comput Electron Agric, 2017,141:351-356.
[11] Amara J, Bouaziz B, Algergawy A. A deep-learning based approach for banana leaf diseases classification. Datenbanksysteme Für Business, 2017,266:79-88.
[12] Dyrmann M, Karstoft H, Midtiby H S. Plant species classification using deep convolutional neural network. Biosyst Eng, 2016,151:72-80.
[13] Ubbens J, Cieslak M, Prusinkiewicz P, Stavness I. The use of plant models in deep learning: an application to leaf counting in rosette plants. Plant Methods, 2018,14:6.
pmid: 29375647
[14] Uzal L C, Grinblat G L, Namías R, Larese M G, Bianchi J S, Morandi E N, Granitto P M. Seed-per-pod estimation for plant breeding using deep learning. Comput Electron Agric, 2018,150:196-204.
[15] Krizhevsky A, Sutskever I, Hinton G. ImageNet classification with deep convolutional neural networks. In: Advances in Neural Information Processing Systems. New York: Curran Associates, 2012. pp 1097-1105.
[16] Karen S, Andrew Z. Very deep convolutional networks for large-scale image recognition. In: Bengio Y, Lecun Y, eds. 3rd International Conference on Learning Representations. San Diego: Conference Track Proceedings, 2015. https://arxiv.org/abs/1409. 1556.
[17] Christian S, Liu W, Jia Y Q, Pierre S, Scott R, Dragomir A, Dumitru E, Vincent V, Andrew R. Going deeper with convolutions. In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston: IEEE Computer Society, 2015. pp 1-9.
[18] He K M, Zhang X Y, Ren S Q, Sun J. Deep residual learning for image recognition. In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE Computer Society, 2016. pp 770-778.
[19] 张德丰. MATLAB数字图像处理. 北京: 机械工业出版社, 2012. pp 276-285.
Zhang D F. MATLAB Digital Image Processing . Beijing: Mechanical Industry Press, 2012. pp 276-285(in Chinese).
[20] 张铮, 倪红霞, 苑春苗, 杨立红. 精通MATLAB数字图像处理与识别. 北京: 人民邮电出版社, 2017. pp 253-289.
Zhang Z, Ni H X, Yuan C M, Yang L H. Proficient in MATLAB Digital Image Processing and Recognition. Beijing: People's Posts and Telecommunications Press, 2017. pp 253-289(in Chinese).
[21] Lempitsky V, Kohli P, Rother C, Sharp T. Image segmentation with a bounding box prior. In: IEEE 12th International Conference on Computer Vision. Kyoto: IEEE Computer Society, 2009. pp 277-284.
[22] Abadi M, Barham P, Chen J M, Chen Z F, Davis A, Dean J, Devin M, Ghemawat S, Irving G, Isard M, Kudlur M, Levenberg J, Monga R, Moore S, Murray D, Steiner B, Tucker P, Vasudevan V, Warden P, Zhang X Q. TensorFlow: a system for large-scale machine learning. In: 12th Symposium on Operating Systems Design and Implementation. Symposium: USENIX Association, 2016. pp 265-283.
[23] Weiss K, Khoshgoftaar T M, Wang D D. A survey of transfer learning. J Big Data, 2016,3:9.
[24] Pan S J, Qiang Y. A survey on transfer learning. IEEE Trans Knowledge Data Eng, 2010,22:1345-1359.
[25] Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S, Huang Z H, Karpathy A, Khosla A, Bernstein M, Berg A C, Li F F. ImageNet large scale visual recognition challenge. Int J Comp Vision, 2015,115:211-252.
[26] Srivastava N, Hinton G, Krizhevsky A, Sutskever I, Salakhutdinov R. Dropout: a simple way to prevent neural networks from overfitting. J Machine Learn Res, 2014,15:1929-1958.
[27] Christian S, Vincent V, Sergey I, Jonathon S, Zbigniew W. Rethinking the inception architecture for computer vision. In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE Computer Society, 2016. pp 2818-2826.
[28] Lin M, Chen Q, Yan S. Network in network. In: Bengio Y, LeCun Y, eds. International Conference on Learning Representations. Banff: Conference Track Proceedings, 2014. https://arxiv.org/abs/ 1312. 4400.
[29] Duchi J, Hazan E, Singer Y. Adaptive subgradient methods for online learning and stochastic optimization. J Machine Learn Res, 2011,12:257-269.
[30] Kingma D, Ba J. Adam: a method for stochastic optimization. In: Bengio Y, LeCun Y, eds. 3rd International Conference on Learning Representations. San Diego: Conference Track Proceedings, 2015. https://arxiv.org/abs/1412.6980v8.
[31] Ng A Y. Feature selection, L1 vs. L2 regularization, and rotational invariance. In: Proceedings of the Twenty-first International Conference on Machine Learning. New York: Association for Computing Machinery, 2004. pp 78-86.
[32] Goodfellow I, Bengio Y, Courville Y. Deep Learning Cambridge: MIT Press, 2016 [2019-11-26]. http://www.deeplearningbook.org.
[33] Ruder S. An overview of gradient descent optimization algorithms. CoRR, 2016,abs/1609.04747. https://arxiv.org/abs/1609. 04747.
[34] Arnal B J G. Impact of dataset size and variety on the effectiveness of deep learning and transfer learning for plant disease classification. Comput Electron Agric, 2018,153:46-53.
doi: 10.1016/j.compag.2018.08.013
[35] Kamilaris A, Prenafeta-Boldú F X. Deep learning in agriculture: a survey. Comp Electr Agric, 2018,147:70-90.
doi: 10.1016/j.compag.2018.02.016
[36] Mehdipour G M, Yanikoglu B, Aptoula E. Plant identification using deep neural networks via optimization of transfer learning parameters. Neurocomputing, 2017,235:228-235.
doi: 10.1016/j.neucom.2017.01.018
[1] 李金敏, 陈秀青, 杨琦, 史良胜. 基于高光谱的水稻叶片氮含量估计的深度森林模型研究[J]. 作物学报, 2021, 47(7): 1342-1350.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!