[1]林志玮,涂伟豪,黄嘉航,等.深度语义分割的无人机图像植被识别[J].山地学报,2018,(06):953-963.[doi:10.16089/j.cnki.1008-2786.000390]
　LIN Zhiwei*,TU Weihao,HUANG Jiahang,et al.Unmanned Aerial Vehicle Vegetation Image Recognition using Deep Semantic Segmentation[J].Mountain Research,2018,(06):953-963.[doi:10.16089/j.cnki.1008-2786.000390]
点击复制

深度语义分割的无人机图像植被识别()
分享到：

《山地学报》[ISSN:1008-2186/CN:51-1516]

卷:
期数:
2018年06期
页码:
953-963
栏目:
山地技术
出版日期:
2018-11-30

文章信息/Info

Title:
Unmanned Aerial Vehicle Vegetation Image Recognition using Deep Semantic Segmentation
文章编号:
1008-2786-(2018)6-953-11
作者:
林志玮1; 2; 3; 涂伟豪1; 黄嘉航2; 丁启禄1; 刘金福1; 4
1.福建农林大学 计算机与信息学院,福建 福州350002; 2.福建农林大学 林学院,福建 福州350002; 3.福建农林大学 林学博士后流动站,福建 福州350002; 4.福建省高校生态与资源统计重点实验室,福建 福州350002
Author(s):
LIN Zhiwei1; 2; 3*;  TU Weihao1;  HUANG Jiahang1;  DING Qilu1;  LIU Jinfu1; 4
1. College of Computer and Information Science, Fujian Agriculture and Forestry University, Fujian Fuzhou 350002, China; 2. College of Forestry, Fujian Agriculture and Forestry University, Fujian Fuzhou 350002, China; 3. Forestry Post-doctoral station of Fujian Agriculture and Forestry University, Fujian Fuzhou 350002, China; 4. Key Laboratory for Ecology and Resource Statistics of Fujian Province, Fujian Fuzhou 350002, China
关键词:
语义分割;  全卷积网络;  无人机影像;  植被识别
Keywords:
semantic segmentation;  fully convolutional networks;  UAV image;  vegetation recognition
分类号:
S719
DOI:
10.16089/j.cnki.1008-2786.000390
文献标志码:
A
摘要:
为有效实施植被信息获取及监测,亟需分类准确及易于推广的植被信息识别技术。本文利用无人机航拍获取植被光学影像,利用深度语义分割技术建构植被种类识别模型,为植被变化动态监测提供准确的植被类别信息。首先,基于安溪县龙门镇崩岗区的采样点,获取20 m航拍高度的无人机影像,构建FCN-VGG19植被识别模型,探讨不同特征融合结构对FCN-VGG19识别性能的影响,测算出各植被的覆盖面积; 其次,取安溪县另一取样点的无人机影像作为验证集,分析FCN-VGG19的迁移学习能力,验证模型稳健性。结果表明:(1)基于20 m高度的无人机影像建立的FCN-VGG19-8s模型识别正确率最高,为86.30%;(2)FCN-VGG19-8s识别精度高于FCN-VGG19-32s; 并从测试集中随机抽取一张图,测算该测试图的马尾松覆盖面积为78.38 m2,芒萁覆盖面积为12.77 m2,柠檬桉覆盖面积为0.89 m2;(3)在模型的迁移学习能力试验分析中,当A数据集占训练集的比例下降时,对模型识别B数据集的影响不大; 当B数据集的数据量减少时,其识别精度稍有下降,仍有84.5%。本文基于无人机光学影像,结合深度语义分割模型进行植被识别,以福建安溪县为例验证模型稳健性,分析模型在测算植被覆盖面积的适用性,旨在为植被识别研究提供新思路。
Abstract:
To efficiently monitor and obtain information on vegetation, a vegetation identification technique is required that yields high classification accuracy and can easily be promoted. To provide precise information on vegetation types to dynamically monitor changes in vegetation, an unmanned aerial vehicle(UAV)was equipped with an optical camera to obtain optical images of vegetation, and the deep semantic segmentation technique was used to construct a model for identifying types of vegetation. First, the FCN-VGG19 vegetation recognition model was constructed on the basis of optical images of a permanent gully captured in Anxi County in China's Fujian Province, to assess the effects of feature fusion with various structures on their performance and calculate the coverage of each vegetation type. Subsequently, optical images captured at another sampling site were used as a validation set to analyze the performance of the FCN-VGG19 model with transfer learning and verify its stability. The experimental results indicated that:(1)the FCN-VGG19-8s model constructed using UAV images at an aerial height of 20 m achieved the highest accuracy of 86.30%.(2)The recognition accuracy of the FCN-VGG19-8s model was higher than that of the FCN-VGG19-32s; an image was randomly extracted from the test dataset, and the calculated coverage of Pinus massoniana Lamb., Miscanthus sinensis Anderss., and Eucalyptus citriodora Hook. f. was 78.38 m2, 12.77 m2, and 0.89 m2, respectively.(3)In the validation of transfer learning, the performance of the model for identifying dataset B was only slightly affected because the proportion of the training data from dataset A was decreased. Although the identification accuracy was slightly decreased when the training data from dataset B were increased, the identification accuracy still reached 84.5%. The optical images obtained from the UAV were associated with a deep semantic segmentation architecture to identify types of vegetation. In addition, Anxi County was used as the sampling area to demonstrate the robustness of the architecture and evaluate its feasibility by calculating the vegetation coverage. Overall, this research aimed to provide an innovative method for vegetation recognition.

参考文献/References:

[1] WILLIS K J. The state of the world's plants report-2017[R]. London, Royal Botanic Gardens, Kew, 2017
[2] 李钰溦,贾坤,魏香琴,等. 中国北方地区植被覆盖度遥感估算及其变化分析[J]. 国土资源遥感,2015,27(2):112-117 [LI Yuwei, JIA Ku, WEI Xiangqing, et al. Fractional vegetation cover estimation in northern China and its change analysis[J]. Remote Sensing for Land and Resources, 2015, 27(2):112 -117]
[3] 丁雷龙,李强子,杜鑫,等. 基于无人机图像颜色指数的植被识别[J]. 国土资源遥感,2016,28(1):78-86 [DING Leilong, LI Qiangzi, DU Xin, et al. Vegetation extraction method based on color indices from UAV images[J]. Remote Sensing for Land and Resources, 2016, 28(1):78-86]
[4] 宋仁飞,林辉,严恩萍,等. 洞庭湖湿地植被分类研究[J]. 中南林业科技大学学报,2015,35(11):27-31 [SONG Renfei, LIN Hui, YAN Enping, et al. Dongting lake wetland vegetation classification[J]. Journal of Central South University of Forestry and Technology, 2015, 35(11):27-31]
[5] 李梦莹,胡勇,王征禹. 基于C5.0决策树和时序HJ-1A/B CCD数据的神农架林区植被分类[J]. 长江流域资源与环境,2016,25(7):1070-1077 [LI Mengying, HU Yong, WANG Zhengyu. Study on vegetation classification in Shennongjia forest district based on C5.0 decision tree and HJ-1A/B data[J]. Resources and Environment in the Yangtze Basin, 2016, 25(7):1070-1077]
[6] 付伟,徐涵秋,王美雅,等. 南方红壤典型水土流失区植被分类及植被类型变化的遥感评估--以福建省长汀县河田地区为例[J]. 遥感技术与应用,2017,32(3):546-555 [FU Wei, XU Hanqiu, WANG Meiya, et al. Vegetation classification and variation assessment in a typical red soil erosion area in southern China: Hetian, Changting County of Fujian Province[J]. Remote Sensing Technology and Application, 2017, 32(3):546-555]
[7] 贾明明,任春颖,刘殿伟,等. 基于环境星与MODIS时序数据的面向对象森林植被分类[J]. 生态学报,2014,34(24):7167-7174 [JIA Mingming, REN Chunyin, LIU Dianwei, et al. Object-oriented forest classification based on combination of HJ-1 CDD and MODIS-NDVI data[J]. Acta Ecologica Sinica, 2014, 34(24):7167-7174]
[8] 刘炜,王聪华,赵尔平,等. 基于分层分级的遥感图像植被分类方法[J]. 中国矿业大学学报,2016,45(4):828-835 [LIU Wei, WANG Conghua, ZHAO Erping, et al. Vegetation classification of remote sensing image based on hierarchical clustering analysis[J]. Journal of China University of Mining and Technology, 2016, 45(4):828-835]
[9] 张华,张改改,吴睿. 基于GF-1卫星数据的面向对象的民勤绿洲植被分类研究[J]. 干旱区地理, 2017,40(4):831-838 [ZHANG Hua, ZHANG Gaigai, WU Rui. Object-based vegetable classification based on GF-1 imagery in Minqin Oasis[J]. Arid Land Geography, 2017, 40(4):831-838]
[10] 费运巧,刘文萍,骆有庆,等. 森林病虫害监测中的无人机图像分割算法比较[J]. 计算机工程与应用,2017,53(8):216-223 [FEI Yunqiao, LIU Wenping, LUO Youqing, et al. Comparison of algorithms for unmanned aerial vehicle image segmentation in monitoring forest diseases and insect pests[J]. Computer Engineering and Applications, 2017, 53(8):216-223]
[11] 雷添杰,李长春,何孝莹. 无人机航空遥感系统在灾害应急救援中的应用[J]. 自然灾害学报,2011,20(1):178-183 [LEI Tianjie, LI Changchun, HE Xiaoying. Application of aerial remote sensing of pilotless aircraft to disaster emergency rescue[J]. Journal of Natural Disasters, 2011, 20(1): 178-183]
[12] KRIZHEVSKY A, SUTSKEVERI, HINTON G E. Imagenet classification with deep convolutional neural networks[C]// Proceeding of the Advances in Neural Information Processing Systems. Harrahs and Harveys, Lake Tahoe, USA: NIPS, 2012: 1097-1105
[13] SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Boston, MA, USA: IEEE, 2015: 1-9
[14] GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Columbus, OH, USA: IEEE, 2014: 580-587
[15] GIRSHICK R. Fast R-CNN [C]// Proceedings of the IEEE International Conference on Computer Vision. Santiago, Chile: IEEE, 2015: 1440-1448
[16] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. Computer Science, 2014
[17] LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Boston, MA, USA: IEEE, 2015: 3431-3440
[18] 林升钰,査轩. 基于GIS的安溪县土壤侵蚀潜在危险度评价[J]. 水土保持研究,2010,17(06):60-63 [LIN Shengyu, CHA Xuan. Study on potential danger of soil erosion based on GIS in Anxi County[J]. Research of Soil and Water Conservation, 2010, 17(06):60-63]
[19] 张学俭,沈雪建. 治理与开发相结合实现生态与经济效益双赢[J]. 中国水土保持,2004(09):5-6 [ZHANG Xuejian, SHEN Xuejian. Realizing win-win of ecological and economic returns through combination of management and development[J]. Soil and Water Conservation in China, 2004(09):5-6]
[20] 易盟,隋立春. 基于改进全卷积神经网络的航拍图像语义分类方法[J]. 计算机工程,2017,43(10):216-221 [YI Meng, SUI Lichun. Aerial image semantic classification method based on improved full convolution neural network[J]. Computer Engineering, 2017, 43(10):216-221]
[21] 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 Linlin, LI Haitao, HAN Yanshun, et al. Application of convolution neural networks in classification of high resolution remote sensing imagery[J]. Science of Surveying and Mapping, 2016, 41(9):170-175]
[22] ZEILER M D, TAYLOR G W, FERGUS R. Adaptive deconvolutional networks for mid and high level feature learning[C]// Proceeding of the IEEE International Conference on Computer Vision. Barcelona, Spain: IEEE, 2011: 2018-2025

备注/Memo

备注/Memo:
收稿日期(Received date):2018-04-24; 改回日期(Accepted date):2018-11-13
基金项目(Foundation item):海峡博士后交流资助计划; 中国博士后科学基金面上项目(2018M632565); 福建省自然科学基金项目(2016J01718)。 [Channel Postdoctoral Exchange Funding Scheme; China Postdoctoral Science Foundation(2018M632565); Natural Science Foundation of Fujian(2016J01718)]
作者简介(Corresponding author):林志玮(1981-),男,博士,讲师,主要研究方向:图像处理,图形识别,机器学习。[LIN Zhiwei(1981-), male, Ph. D., lecturer, research on image processing, pattern recognition and machine learning] E-mail: cwlin@fafu.edu.cn
更新日期/Last Update: 2018-11-30