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

基于卷积神经网络的高分二号影像林分类型分类

江涛 王新杰

江涛, 王新杰. 基于卷积神经网络的高分二号影像林分类型分类[J]. 北京林业大学学报, 2019, 41(9): 20-29. doi: 10.13332/j.1000-1522.20180342
引用本文: 江涛, 王新杰. 基于卷积神经网络的高分二号影像林分类型分类[J]. 北京林业大学学报, 2019, 41(9): 20-29. doi: 10.13332/j.1000-1522.20180342
Jiang Tao, Wang Xinjie. Convolutional neural network for GF-2 image stand type classification[J]. Journal of Beijing Forestry University, 2019, 41(9): 20-29. doi: 10.13332/j.1000-1522.20180342
Citation: Jiang Tao, Wang Xinjie. Convolutional neural network for GF-2 image stand type classification[J]. Journal of Beijing Forestry University, 2019, 41(9): 20-29. doi: 10.13332/j.1000-1522.20180342

基于卷积神经网络的高分二号影像林分类型分类

doi: 10.13332/j.1000-1522.20180342
基金项目: 国家重点研发计划项目(2017YFC050410101),中央高校基本科研业务费专项(BLJD200907、JD2010-2)
详细信息
    作者简介:

    江涛。主要研究方向:林业遥感与信息技术。Email:taoorwell@bjfu.edu.cn  地址:100083 北京市海淀区清华东路35号北京林业大学林学院

    责任作者:

    王新杰,教授。主要研究方向:森林资源监测和林业遥感。Email:xinjiew@bjfu.edu.cn  地址:同上

  • 中图分类号: S771.8

Convolutional neural network for GF-2 image stand type classification

  • 摘要: 目的基于遥感影像的林分类型分类在现代林业中是一项重要的应用。本文试图构建一个基于高分二号(GF-2)影像林分类型分类的卷积神经网络(CNN)模型,探索CNN在遥感图像像素级分类这一领域的发展潜力。方法以GF-2卫星遥感影像为数据源,利用Tensorflow(一种开源用于机器学习的框架)构建4种不同图像斑块大小(m = 5,7,9,11)为输入的CNN,同时以传统的神经网络模型——多层感知器(MLP)为基准,比较不同图像斑块大小下的CNN分类图的分类效果和分类精度。结果实验分类结果表明:CNN(m = 9)得出最高的分类精确度,总体精度比MLP和CNN(m = 5,7,11)分别高出10.91%和6.55%、1.3%、2.54%。分类图的可视化结果也表明CNN(m = 9)更好地解决了“椒盐现象”与过度平滑后的边界不确定性问题。结论CNN能够在利用高分影像光谱特征的同时充分挖掘影像的空间特征,从而提高分类精度,同时在利用CNN基于遥感影像分类时,根据数据源以及地物的特点选择合适的图像斑块大小作为输入是提高分类精度与分类效果的关键措施。

     

  • 图  1  MLP和CNN(m = 9)的结构框架

    Figure  1.  Architecture of the CNN (m = 9) and MLP

    图  2  CNN(m = 9)在训练阶段准确率与损失值图

    Figure  2.  Accuracy (left) and loss (right) of the CNN (m = 9) in stage of training

    图  3  两幅研究区子图在不同分类系统下的分类结果

    Figure  3.  Two image subsets (A and B) in study area with classification results (Columns from left to right represent the original images, the MLP classification and CNN (m = 5, 7, 9, 11) classification correspondingly)

    表  1  研究区分类类别描述及其训练、验证和测试样本大小

    Table  1.   Description of class at study with training, validation and testing samples size

    林分类型
    Stand type
    详细描述
    Descriptions
    训练样本
    Train sample
    验证样本
    Validation sample
    测试样本
    Test sample
    灌草
    Shrub
    林中草地、生长大量灌草的伐后林地、布满杂草的林道
    Grass glade, nonforested land with large amount bushes and grasses,
    forest road with large shrubs
    1 653 525 690
    马尾松
    Masson pine
    不同龄级的马尾松林,主要是近熟林和成熟林
    Masson pine forest with different age classes,
    mainly half-mature forest and mature forest
    2 008 661 1 289
    其他阔叶林
    Other broadleaf
    一些阔叶树,主要是由木荷组成的林带、
    散生的泡桐以及其他阔叶树种组成的混交林
    Some broadleaf trees, including schima used as fire belt,
    scattered paulownia and other mixed broadleaf forest
    1 823 590 1 190
    裸地
    Bare soil
    主要是几乎不含植被的农业用地和林业用地
    Cropland and forest land with low percentage or no vegetation
    1 653 575 347
    杉木林
    Chinese fir
    不同龄级的杉木林,主要是幼龄林、中龄林和成熟林
    Chinese fir forest with different age classes, including young forest,
    half-mature forest and mature forest
    2 743 943 1 548
    下载: 导出CSV

    表  3  基于CNN生成的混淆矩阵(m = 5)

    Table  3.   Confusion matrix using the CNN (m = 5)

    林分类型 Stand type地面真实点 Ground truth (pixel)
    灌草 Shrub马尾松林 Masson pine其他阔叶林 Other broadleaf裸地 Bare soil杉木林 Chinese fir
    灌草 Shrub619 0300 0 0
    马尾松林 Masson pine 01 190 0 0 53
    其他阔叶林 Other broadleaf 67 0873 0 1
    裸地 Bare soil 0 0 0337 0
    杉木林 Chinese fir 4 131 17 101 350
    生产者精度 Producer’s accuracy/% 89.71 90.83 73.36 97.12 96.15
    用户精度 User’s accuracy/% 67.36 95.74 92.77100 89.29
    下载: 导出CSV

    表  4  基于CNN生成的混淆矩阵(m = 7)

    Table  4.   Confusion matrix using the CNN (m = 7)

    林分类型 Stand type地面真实点 Ground truth (pixel)
    灌草 Shrub马尾松林 Masson pine其他阔叶林 Other broadleaf裸地 Bare soil杉木林 Chinese fir
    灌草 Shrub620 0 126 0 0
    马尾松林 Masson pine 01 263 0 0 9
    其他阔叶林 Other broadleaf 69 01 064 0 1
    裸地 Bare soil 0 0 0325 0
    杉木林 Chinese fir 1 58 0 221 647
    生产者精度 Producer’s accuracy/% 89.86 95.61 89.41 93.66 96.65
    用户精度 User’s accuracy/% 83.11 97.45 92.76100 94.37
    下载: 导出CSV

    表  5  基于CNN生成的混淆矩阵(m = 9)

    Table  5.   Confusion matrix using the CNN (m = 9)

    林分类型 Stand type地面真实点 Ground truth (pixel)
    灌草 Shrub马尾松林 Masson pine其他阔叶林 Other broadleaf裸地 Bare soil杉木林 Chinese fir
    灌草 Shrub605 0 59 0 0
    马尾松林 Masson pine 01 288 0 0 30
    其他阔叶林 Other broadleaf 78 111 095 0 19
    裸地 Bare soil 0 0 0342 0
    杉木林 Chinese fir 7 13 36 51 355
    生产者精度 Producer’s accuracy/% 87.68 98.17 92.02 98.56 96.51
    用户精度 User’s accuracy/% 91.11 97.72 91.02100 95.69
    下载: 导出CSV

    表  2  基于MLP生成的混淆矩阵

    Table  2.   Confusion matrix using the MLP

    林分类型 Stand type地面真实点 Ground truth (pixel)
    灌草 Shrub马尾松林 Masson pine其他阔叶林 Other broadleaf裸地 Bare soil杉木林 Chinese fir
    灌草 Shrub515 0340 0 0
    马尾松林 Masson pine 01 200 0 0 130
    其他阔叶林 Other broadleaf169 0844 0 20
    裸地 Bare soil 0 0 0342 2
    杉木林 Chinese fir 6 121 6 51 252
    生产者精度 Producer’s accuracy/% 74.64 90.84 70.92 98.56 89.17
    用户精度 User’s accuracy/% 60.23 90.23 81.70 99.42 90.07
    下载: 导出CSV

    表  6  基于CNN生成的混淆矩阵

    Table  6.   Confusion matrix using the CNN (m = 11)

    林分类型 Stand type地面真实点 Ground truth (pixel)
    灌草 Shrub马尾松林 Masson pine其他阔叶林 Other broadleaf裸地 Bare soil杉木林 Chinese fir
    灌草 Shrub531 0 1 0 0
    马尾松林 Masson pine 01 262 0 0 52
    其他阔叶林 Other broadleaf120 41 137 0 24
    裸地 Bare soil 0 0 0293 0
    杉木林 Chinese fir 39 37 52 541 328
    生产者精度 Producer’s accuracy/% 76.96 96.85 95.55 84.44 94.59
    用户精度 User’s accuracy/% 99.81 96.04 88.48100 87.95
    下载: 导出CSV

    表  7  不同分类系统分类精度对比利用每种类别的Fscore、OA和Kappa

    Table  7.   Classification accuracy comparison amongst MLP, CNN (m = 5), CNN (m = 7), CNN (m = 9) and CNN (m = 11) for study area using per-class F score, overall accuracy and kappa coefficient (The bold font highlights the greatest classification accuracy per row)

    林分类型 Stand type分类器 Classifier
    MLPCNN (m = 5)CNN (m = 7)CNN (m = 9)CNN (m = 11)
    灌草 Shrub0.666 70.769 40.863 50.893 60.869 1
    马尾松林 Masson pine0.905 30.928 20.965 20.979 40.964 4
    其他阔叶林 Other broadleaf0.759 30.819 30.910 60.915 20.918 8
    裸地 Bare soil0.989 90.985 40.967 30.992 70.915 6
    杉木林 Chinese fir0.896 20.925 90.955 00.961 00.911 5
    总体精度 Overall accuracy/%83.8788.2393.4894.7892.24
    Kappa 系数 Kappa coefficient0.790 40.847 10.914 90.931 80.897 9
    下载: 导出CSV
  • [1] 樊文锋, 李鸿洲, 温奇, 等. 高分二号卫星影像正射纠正精度分析[J]. 测绘通报, 2016(9):63−66.

    Fan W F, Li H Z, Wen Q, et al. Research on accuracy of digital Ortho Rectification of GF-2 image[J]. Bulletin of Surveying and Mapping, 2016(9): 63−66.
    [2] 陈业培, 孙开敏, 白婷, 等. 高分二号影像融合方法质量评价[J]. 测绘科学, 2017,42(11):35−40.

    Chen Y P, Sun K M, Bai T, et al. Quality assessment on image fusion methods for GF-2 data[J]. Science of Surveying and Mapping, 2017,42(11): 35−40.
    [3] Mountrakis G, Im J, Ogole C. Support vector machines in remote sensing: a review[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2011, 66(3): 247−259. doi: 10.1016/j.isprsjprs.2010.11.001
    [4] Rodriguez-Galiano V F, Chica-Olmo M, Abarca-Hernandez F, et al. Random forest classification of Mediterranean land cover using multi-seasonal imagery and multi-seasonal texture[J]. Remote Sensing of Environment, 2012, 121: 93−107. doi: 10.1016/j.rse.2011.12.003
    [5] Mas J F, Flores J J. The application of artificial neural networks to the analysis of remotely sensed data[J]. International Journal of Remote Sensing, 2008, 29(3): 617−663. doi: 10.1080/01431160701352154
    [6] Kumar U, Kerle N, Punia M, et al. Mining land cover information using multilayer perceptron and decision tree from MODIS data[J]. Journal of the Indian Society of Remote Sensing, 2010, 38(4): 592−603. doi: 10.1007/s12524-011-0061-y
    [7] Hu F, Xia G, Hu J, et al. Transferring deep convolutional neural networks for the scene classification of high-resolution remote sensing imagery[J]. Remote Sensing, 2015, 7(11): 14680−14707. doi: 10.3390/rs71114680
    [8] 葛芸, 江顺亮, 叶发茂, 等. 基于ImageNet预训练卷积神经网络的遥感图像检索[J]. 武汉大学学报(信息科学版), 2018, 43(1):67−73.

    Ge Y, Jiang S L, Ye F M, et al. Remote sensing image retrieval using pre-trained convolutional neural networks based on ImageNet[J]. Geomatics and Information Science of Wuhan University, 2018, 43(1): 67−73.
    [9] 姚相坤, 万里红, 霍宏, 等. 基于多结构卷积神经网络的高分遥感影像飞机目标检测[J]. 计算机工程, 2017,43(1):259−267. doi: 10.3969/j.issn.1000-3428.2017.01.045

    Yao X K, Wan L H, Huo H, et al. Airplane object detection in high resolution remote sensing imagery based on multi-structure convolutional neural network[J]. Computer Engineering, 2017,43(1): 259−267. doi: 10.3969/j.issn.1000-3428.2017.01.045
    [10] 黄洁, 姜志国, 张浩鹏, 等. 基于卷积神经网络的遥感图像舰船目标检测[J]. 北京航空航天大学学报, 2017,43(9):1841−1848.

    Huang J, Jiang Z G, Zhang H P, et al. Ship object detection in remote sensing images using convolutional neural networks[J]. Journal of Beijing University of Aeronautics and Astronautics, 2017,43(9): 1841−1848.
    [11] Ren Y, Zhu C, Xiao S. Small object detection in optical remote sensing images via modified faster R-CNN[J]. Applied Sciences, 2018, 8(5): 813. doi: 10.3390/app8050813
    [12] Sharma A, Liu X, Yang X, et al. A patch-based convolutional neural network for remote sensing image classification[J]. Neural Networks, 2017, 95: 19−28. doi: 10.1016/j.neunet.2017.07.017
    [13] 付秀丽, 黎玲萍, 毛克彪, 等. 基于卷积神经网络模型的遥感图像分类[J]. 高技术通讯, 2017,27(3):203−212. doi: 10.3772/j.issn.1002-0470.2017.03.002

    Fu X L, Li L P, Mao K B, et al. Remote sensing image classification based on CNN model[J]. Chinese High Technology Letters, 2017,27(3): 203−212. doi: 10.3772/j.issn.1002-0470.2017.03.002
    [14] Zhang C, Pan X, Li H, et al. A hybrid MLP-CNN classifier for very fine resolution remotely sensed image classification[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 140: 133−144. doi: 10.1016/j.isprsjprs.2017.07.014
    [15] 曹林林, 李海涛, 韩颜顺, 等. 卷积神经网络在高分遥感影像分类中的应用[J]. 测绘科学, 2016,41(9):170−175.

    Cao L L, Li H T, Han Y S, et al. Application of convolutional neural networks in classification of high resolution remote sensing imagery[J]. Science of Surveying and Mapping, 2016,41(9): 170−175.
    [16] 陈广胜, 陈守玉, 景维鹏, 等. 基于遗传算法与全卷积网络的遥感影像像素级分类方法[J]. 小型微型计算机系统, 2018,39(7):1590−1595. doi: 10.3969/j.issn.1000-1220.2018.07.041

    Chen G S, Chen S Y, Jing W P, et al. Remote sensing image pixel-level classification with genetic algorithm and fully convolutional network[J]. Journal of Chinese Computer Systems, 2018,39(7): 1590−1595. doi: 10.3969/j.issn.1000-1220.2018.07.041
    [17] Yue J, Mao S, Li M. A deep learning framework for hyperspectral image classification using spatial pyramid pooling[J]. Remote Sensing Letters, 2016, 7(9): 875−884. doi: 10.1080/2150704X.2016.1193793
    [18] 李竺强, 朱瑞飞, 高放, 等. 三维卷积神经网络模型联合条件随机场优化的高光谱遥感影像分类[J]. 光学学报, 2018,38(8):404−413.

    Li Z Q, Zhu R F, Gao F, et al. Hyperspectral remote sensing image classification based on three-dimensional convolution neural network combined with conditional random field optimization[J]. Act Optica Sinica, 2018,38(8): 404−413.
    [19] Längkvist M, Kiselev A, Alirezaie M, et al. Classification and segmentation of satellite orthoimagery using convolutional neural networks[J]. Remote Sensing, 2016, 8(4): 329. doi: 10.3390/rs8040329
    [20] Fassnacht F E, Latifi H, Stereńczak K, et al. Review of studies on tree species classification from remotely sensed data[J]. Remote Sensing of Environment, 2016, 186: 64−87. doi: 10.1016/j.rse.2016.08.013
    [21] Dalponte M, Bruzzone L, Gianelle D. Tree species classification in the Southern Alps based on the fusion of very high geometrical resolution multispectral/hyperspectral images and LiDAR data[J]. Remote Sensing of Environment, 2012, 123: 258−270. doi: 10.1016/j.rse.2012.03.013
    [22] 王妮, 彭世揆, 李明诗. 基于树种分类的高分辨率遥感数据纹理特征分析[J]. 浙江农林大学学报, 2012,29(2):210−217. doi: 10.3969/j.issn.2095-0756.2012.02.010

    Wang N, Peng S K, Li M S. High-resolution remote sensing of texture images for tree species classification[J]. Journal of Zhejiang A & F University, 2012,29(2): 210−217. doi: 10.3969/j.issn.2095-0756.2012.02.010
    [23] 傅锋, 王新杰, 汪锦, 等. 高分二号影像树种识别及龄组划分研究[J]. 国土资源遥感, 2019, 31(2):118−124.

    Fu F, Wang X J, Wang J, et al. Tree species and age groups classification based on GF-2 image[J]. Remote Sensing for Land and Resources, 2019, 31(2): 118−124.
    [24] 李丹, 柯樱海, 宫辉力, 等. 基于高分辨率遥感影像的城市典型乔木树种分类研究[J]. 地理与地理信息科学, 2016,32(1):84−89. doi: 10.3969/j.issn.1672-0504.2016.01.016

    Li D, Ke Y H, Gong H L, et al. Urban tree species classification with machine learning classifier using WorldView-2 imagery[J]. Geography and Geo-Information Science, 2016,32(1): 84−89. doi: 10.3969/j.issn.1672-0504.2016.01.016
  • 加载中
图(3) / 表(7)
计量
  • 文章访问数:  2178
  • HTML全文浏览量:  1587
  • PDF下载量:  94
  • 被引次数: 0
出版历程
  • 收稿日期:  2018-10-25
  • 修回日期:  2019-03-15
  • 网络出版日期:  2019-08-07
  • 刊出日期:  2019-09-01

目录

    /

    返回文章
    返回