高级检索

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于深度学习的5种树皮纹理图像识别研究

刘嘉政 王雪峰 王甜

刘嘉政, 王雪峰, 王甜. 基于深度学习的5种树皮纹理图像识别研究[J]. 北京林业大学学报, 2019, 41(4): 146-154. doi: 10.13332/j.1000-1522.20180242
引用本文: 刘嘉政, 王雪峰, 王甜. 基于深度学习的5种树皮纹理图像识别研究[J]. 北京林业大学学报, 2019, 41(4): 146-154. doi: 10.13332/j.1000-1522.20180242
Liu Jiazheng, Wang Xuefeng, Wang Tian. Research on image recognition of five bark texture images based on deep learning[J]. Journal of Beijing Forestry University, 2019, 41(4): 146-154. doi: 10.13332/j.1000-1522.20180242
Citation: Liu Jiazheng, Wang Xuefeng, Wang Tian. Research on image recognition of five bark texture images based on deep learning[J]. Journal of Beijing Forestry University, 2019, 41(4): 146-154. doi: 10.13332/j.1000-1522.20180242

基于深度学习的5种树皮纹理图像识别研究

doi: 10.13332/j.1000-1522.20180242
基金项目: 国家重点研发计划项目(2017YFC0504106)
详细信息
    作者简介:

    刘嘉政。主要研究方向:林业信息技术应用。Email:liujiazheng0919@163.com 地址:100091 北京市海淀区香山路东小府1号中国林业科学研究院资源信息研究所

    责任作者:

    王雪峰,博士,研究员。主要研究方向:森林资源监测与计算机视觉。Email:xuefeng@ifrit.ac.cn 地址:同上

  • 中图分类号: S757.3;TP391

Research on image recognition of five bark texture images based on deep learning

  • 摘要: 目的针对在树皮图像识别时,现有的算法和识别过程过于复杂的问题,提出了基于深度学习的方法来对不同树种的树皮图像进行识别。方法本文以5种常见树种的树皮纹理图像为例,采用基于卷积神经网络的深度学习方法,将原始图像直接作为输入,通过卷积和池化层对图像的低级、高级特征进行自动提取,解决了手动提取纹理特征的困难和问题;在此基础上,对CNN模型结构进行改进,采用带Maxout的ELU激励函数来代替ReLU函数,解决模型的偏移和零梯度问题;对损失函数进行改进,通过添加规范项来优化结构参数,并使用分段常数衰减法对学习率进行动态调控;最后采用softmax分类器对图像类别进行输出。结果对5个树种的树皮图像共计10 000张图像进行实验,其中每类选取200张图像作为测试集。最终训练准确率达到93.80%,测试集识别准确率为97.70%。另外,为验证本文方法的可行性,与传统人工特征提取法,提取HOG特征、Gabor特征和灰度共生矩阵统计法,训练SVM分类器。通过实验比较,本文方法识别准确率最高。结论本文提出的基于深度学习的树皮纹理图像识别方法是可行的,提高了识别效率和精度,为树种的智能化识别提供新的参考。

     

  • 图  1  原始图像和感兴趣区提取

    Figure  1.  Original image and region of interest extraction

    图  2  5个树种树皮图像示例

    Figure  2.  Example of 5 kinds of tree bark images

    图  3  本文实验CNN网络结构图

    Input. 输入层Input layer;Conv. 卷积层Convolutional layer;Pooling. 池化层Pooling layer;LRN. 局部响应归一化Local response normalization;ELU. 指数线性单元Exponential linear unit;FC. 全连接层Fully connected layer;softmax. 分类器Classifier

    Figure  3.  Experimental CNN network structure diagram

    图  4  不同卷积核大小对实验结果的影响

    Figure  4.  Effects of different convolution kernel sizes on experimental results

    图  5  分段常数衰减法调控学习率

    Figure  5.  Piecewise constant attenuation method to control learning rate

    图  6  采用不同激活函数的实验结果

    Figure  6.  Experimental results with different activation functions

    图  7  特征图可视化

        a. 原始图像 Original image;b. 卷积特征图 Convolution map;c. HOG特征和Gabor特征 HOG features and Gabor features

    Figure  7.  Feature map visualization

    图  8  部分误判的树皮图像

    a、b. 落叶松被误判为雪松Larix gmelinii was misidentified as Cedrus deodara;c. 落叶松被误判为山杨Larix gmelinii was misidentified as Populus davidiana;d. 山杨被误判为雪松Populus davidiana was misidentified as Cedrus deodara;e、f. 山杨被误判为白桦Populus davidiana was misidentified as Betula platyphylla

    Figure  8.  Some misidentified bark images

    表  1  实验结果

    Table  1.   Experimental results %

    树种 Tree species 测试准确率 Test accuracy rate
    落叶松 Larix gmelinii 96.5
    白桦 Betula platyphylla 99.5
    山杨 Populus davidiana 97.5
    雪松 Cedrus deodara 95.0
    白皮松 Pinus bungeana 100 
    平均值 Mean 97.70
    下载: 导出CSV

    表  2  5个树种树皮纹理图像判别结果的混淆矩阵

    Table  2.   Confusion matrix of discrimination results of bark texture images of 5 tree species

    树种
    Tree species
    落叶松
    Larix gmelinii
    白桦
    Betula platyphylla
    山杨
    Populus davidiana
    雪松
    Cedrus deodara
    白皮松
    Pinus bungeana
    落叶松
    Larix gmelinii
    193 0 3 4 0
    白桦
    Betula platyphylla
    1 199 0 0 0
    山杨
    Populus davidiana
    0 1 195 4 0
    雪松
    Cedrus deodara
    6 0 4 190 0
    白皮松
    Pinus bungeana
    0 0 0 0 200
    下载: 导出CSV

    表  3  4种方法的识别准确率与识别时间

    Table  3.   Identification accuracy rate and recognition time of four methods

    方法
    Method
    识别准确率
    Recognition
    accuracy rate/%
    识别时间
    Recognition time/s
    灰度共生矩阵
    GLCM
    48.61 4.5
    HOG + SVM 65.24 3.6
    Gabor + SVM 82.35 3.5
    本文方法
    Method used in this paper
    97.70 1.7
    下载: 导出CSV
  • [1] Wooten J R, To S D F, Igathinathane C, et al. Discrimination of bark from wood chips through texture analysis by image processing[J]. Computers and Electronics in Agriculture, 2011, 79(1): 13−19. doi: 10.1016/j.compag.2011.08.005
    [2] Bertrand S, Ameur R B, Cerutti G, et al. Bark and leaf fusion systems to improve automatic tree species recognition[J]. Ecological Informatics, 2018, 46: 57−73. doi: 10.1016/j.ecoinf.2018.05.007
    [3] Kamal K, Qayyum R, Mathavan S, et al. Wood defects classification using laws texture energy measures and supervised learning approach[J]. Advanced Engineering Informatics, 2017, 34: 125−135. doi: 10.1016/j.aei.2017.09.007
    [4] 多化豫, 高峰, 李福胜, 等. 基于图像处理的木片与树皮的新识别参数研究[J]. 西北林学院学报, 2015, 30(1):207−210. doi: 10.3969/j.issn.1001-7461.2015.01.34

    Duo H Y, Gao F, Li F S, et al. Approaching to the new identification parameter on wood and bark based on image processing[J]. Journal of Northwest Forestry University, 2015, 30(1): 207−210. doi: 10.3969/j.issn.1001-7461.2015.01.34
    [5] 赵作林. 基于图像分析的北京地区杨树种类识别研究[D]. 北京: 北京林业大学, 2015.

    Zhao Z L. Research on identification of poplar species in Beijing area based on image analysis[D]. Beijing: Beijing Forestry University, 2015.
    [6] 孙伶君, 汪杭军, 祁亨年. 基于分块LBP的树种识别研究[J]. 北京林业大学学报, 2011, 33(4):107−112.

    Sun L J, Wang H J, Qi H N. Wood recognition based on block local binary pattern (LBP)[J]. Journal of Beijing Forestry University, 2011, 33(4): 107−112.
    [7] 李可心, 戚大伟, 牟洪波, 等. 基于灰度共生矩阵与SOM神经网络的树皮纹理特征识别[J]. 森林工程, 2017, 33(3):24−27. doi: 10.3969/j.issn.1006-8023.2017.03.006

    Li K X, Qi D W, Mou H B, et al. Identification of tree bark texture characteristic based on gray co-occurrence matrix and SOM Neural Network[J]. Forest Engineering, 2017, 33(3): 24−27. doi: 10.3969/j.issn.1006-8023.2017.03.006
    [8] Zeiler M D, Fergus R. Visualizing and understanding convolutional networks[C]//Proceedings of the 13th European Conference on Computer Vision. Zurich: Springer, 2014: 818−833.
    [9] Szegedy C, Liu W, Jia Y Q, et al. Going deeper with convolutions[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston: IEEE, 2015: 1−9.
    [10] Hubel D H, Wiesel T N. Receptive fields, binocular interaction and functional architecture in the cat ’s visual cortex[J]. The Journal of Physiology, 1962, 160(1): 106−154. doi: 10.1113/jphysiol.1962.sp006837
    [11] Lecun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278−2324. doi: 10.1109/5.726791
    [12] 周飞燕, 金林鹏, 董军. 卷积神经网络研究综述[J]. 计算机学报, 2017, 40(6):1229−1251.

    Zhou F Y, Jin L P, Dong J. Review of convolutional neural network[J]. Chinese Journal of Computers, 2017, 40(6): 1229−1251.
    [13] Boureau Y L, Bach F, Lecun Y, et al. Learning mid-level features for recognition[C]//Proceedings of 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Francisco: IEEE, 2010.
    [14] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J/OL]. arXiv, 2015[2018−05−03]. https://arxiv.org/pdf/1409.1556.pdf.
    [15] Yoo H J. Deep convolution neural networks in computer vision[J]. IEIE Transactions on Smart Processing & Computing, 2015, 4(1): 35−43.
    [16] 蒋宗礼. 人工神经网络导论[M]. 北京: 高等教育出版社, 2001.

    Jiang Z L.Introduction to artificial neural networks[M]. Beijing: Higher Education Press, 2001.
    [17] Glorot X, Bordes A, Bengio Y. Deep sparse rectifier neural networks[C/OL]//Proceedings of the 14th International Conference on Artificial Intelligence and Statistics[2018−03−12]. https://www.researchgate.net/publication/215616967_Deep_Sparse_Rectifier_Neural_Networks.
    [18] Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]//International Conference on Computer Vision & Pattern Recognition (CVPR ’05). San Diego: IEEE Computer Society, 2005: 886−893.
    [19] Lee T S. Image representation using 2D Gabor wavelets[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1996, 18(10): 959−971. doi: 10.1109/34.541406
    [20] Haralick R M, Shanmugam K. Textural features for image classification[J]. IEEE Transactions on Systems, Man, and Cybernetics, 1973, SMC-3(6): 610−621.
    [21] Hearst M A, Dumais S T, Osuna E, et al. Support vector machines[J]. IEEE Intelligent Systems and Their Applications, 1998, 13(4): 18−28. doi: 10.1109/5254.708428
    [22] Bahrampour S, Ramakrishnan N, Schott L, et al. Comparative study of deep learning software frameworks[J/OL]. arXiv, 2016 [2018−03−05]. https://arxiv.org/pdf/1511.06435.pdf.
  • 加载中
图(8) / 表(3)
计量
  • 文章访问数:  3082
  • HTML全文浏览量:  2389
  • PDF下载量:  125
  • 被引次数: 0
出版历程
  • 收稿日期:  2018-07-25
  • 修回日期:  2019-03-06
  • 网络出版日期:  2019-04-02
  • 刊出日期:  2019-04-01

目录

    /

    返回文章
    返回