Vegetation extraction algorithm for the Tibetan Plateau based on YOLOv5 and improved DeeplabV3+

doi:10.11686/cyxb2024060

Abstract

Abstract:

Vegetation coverage on the Qinghai-Tibet Plateau is a crucial metric for ecological studies and environmental monitoring. Traditional methods to detect vegetation coverage are effective in regions with simple terrains and concentrated vegetation. However， in complex terrains， issues such as high costs， restricted survey areas， and extended time intervals reduce the accuracy of the results obtained using such traditional methods. In recent years， rapid advancements in computer vision and deep learning have created new opportunities for precise vegetation extraction in the complex terrains of the Qinghai-Tibet Plateau. Here， we introduce a two-stage vegetation extraction algorithm that integrates YOLOv5 and an improved DeeplabV3+. The algorithm utilizes a vegetation detection model based on YOLOv5 to minimize background interference during the second stage of vegetation segmentation； and a newly designed DeeplabV3+ semantic segmentation model for accurate vegetation segmentation and extraction. The improved model incorporates the lightweight backbone network MobileNetV2， optimizes the dilated convolution parameters of the ASPP module， and integrates EMA and CloAttention mechanisms. The experimental results on the unmanned aerial vehicle dataset of the Qinghai-Tibet Plateau demonstrate that the algorithm attains an intersection over union （IoU） of 90.40% and a pixel accuracy （PA） of 96.32%， significantly outperforming other current technologies and greatly reducing the model’s parameters. Under various environmental conditions， the algorithm exhibits high-precision capabilities for vegetation extraction， offering effective technical support for the rapid and precise measurement of vegetation cover on the Qinghai-Tibet Plateau.

Key words: Tibetan Plateau, vegetation extraction, deep learning, YOLOv5, DeeplabV3+

Chu-qi YAN, Jian-qiang HUANG. Vegetation extraction algorithm for the Tibetan Plateau based on YOLOv5 and improved DeeplabV3+[J]. Acta Prataculturae Sinica, 2025, 34(1): 41-54.

Figures/Tables 12

References 33

1	Dong S. Revitalizing the grassland on the Qinghai-Tibetan Plateau. Grassland Research， 2023， 2（3）： 241-250.
2	Sang J W， Song C Y， Jia N X， et al. Vegetation survey and mapping on the Qinghai-Tibet Plateau. Biodiversity Science， 2023， 31（3）： 56-71.
	桑佳文，宋创业，贾宁霞，等. 青藏高原植被调查与制图评估. 生物多样性， 2023， 31（3）： 56-71.
3	Zhang W B， Fu S H， Liu B Y. Error assessment of visual estimation plant coverage. Journal of Beijing Normal University （Natural Science）， 2001， 37（3）： 402-408.
	章文波，符素华，刘宝元. 目估法测量植被覆盖度的精度分析. 北京师范大学学报（自然科学版）， 2001， 37（3）： 402-408.
4	Zhang W B， Liu B Y， Wu J D. Monitoring of plant coverage of plots by visual estimation and overhead photograph. Bulletin of Soil and Water Conservation， 2001， 21（6）： 60-63.
	章文波，刘宝元，吴敬东. 小区植被覆盖度动态快速测量方法研究. 水土保持通报， 2001， 21（6）： 60-63.
5	Yang Q， Pu H M， Zhao X C， et al. Comparison of different plant cover investigation methods for three artificial grasslands. Chinese Journal of Applied & Environmental Biology， 2021， 27（1）： 220-227.
	杨琴，蒲红梅，赵学春，等. 3种人工草地不同植被覆盖度实地测量方法比较. 应用与环境生物学报， 2021， 27（1）： 220-227.
6	Canfield R H. Application of the line intercept method in sampling range vegetation. Journal of Forestry， 1941， 39（4）： 388-394.
7	Zhang Y X， Li X B， Chen Y H. Overview of field and multi-scale remote sensing measurement approaches to grassland vegetation coverage. Advances in Earth Science， 2003， 18（1）： 85-93.
	张云霞，李晓兵，陈云浩. 草地植被盖度的多尺度遥感与实地测量方法综述. 地球科学进展， 2003， 18（1）： 85-93.
8	Huang P， Pu J W， Zhao Q Q， et al. Research progress and development trend of remote sensing information extraction methods of vegetation. Remote Sensing for Natural Resources， 2022， 34（2）： 10-19.
	黄佩，普军伟，赵巧巧，等. 植被遥感信息提取方法研究进展及发展趋势. 自然资源遥感， 2022， 34（2）： 10-19.
9	Shen M X， He R Y， Cong J H， et al. Study on extraction of vegetation information of ETM+ by using PCA method and Brovey transform. Transactions of the Chinese Society for Agricultural Machinery， 2007， 38（9）： 87-89.
	沈明霞，何瑞银，丛静华，等. 基于主成分分析与Brovey变换的ETM+影像植被信息提取. 农业机械学报， 2007， 38（9）： 87-89.
10	Tang P Q， Wu W B， Yao Y M， et al. New method for extracting multiple cropping index of North China Plain based on wavelet transform. Transactions of the Chinese Society of Agricultural Engineering， 2011， 27（7）： 220-225.
	唐鹏钦，吴文斌，姚艳敏，等. 基于小波变换的华北平原耕地复种指数提取. 农业工程学报， 2011， 27（7）： 220-225.
11	Zhang X Y， Jing Y S， Li W G. Optimal scale screening of paddy rice in remote sensing imagery based on high pass filter fusion. Chinese Journal of Agrometeorology， 2018， 39（5）： 344-353.
	张晓忆，景元书，李卫国. 基于高通滤波算法的水稻遥感影像适宜尺度筛选. 中国农业气象， 2018， 39（5）： 344-353.
12	Dai P Q， Ding L X， Liu L J， et al. Tree species identification based on FCN using the visible images obtained from an unmanned aerial vehicle. Laser & Optoelectronics Progress， 2020， 57（10）： 101001.
	戴鹏钦，丁丽霞，刘丽娟，等. 基于FCN的无人机可见光影像树种分类. 激光与光电子学进展， 2020， 57（10）： 101001.
13	Redmon J， Divvals S， Grishick R， et al. You Only Look Once： unified， real time object detection//Institute of Electrical and Electronic Engineers. Conference on Computer Vision and Pattern Recognition. Las Vegas： Institute of Electrical and Electronic Engineers， 2016： 779-788.
14	Krizhevsky A， Sutskever I， Hinton G E. Imagenet classification with deep convolutional neural networks. Communications of the ACM， 2017， 60（6）： 84-90.
15	Chen L C， Papandreou G， Kokkinos I， et al. Semantic image segmentation with deep convolutional nets and fully connected CRFs. Computer Science， 2014（4）： 357-361.
16	Chen L C， Papandreou G， Kokkinos I， et al. Deeplab： Semantic image segmentation with deep convolutional nets， atrous convolution， and fully connected crfs. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2017， 40（4）： 834-848.
17	Chen L C， Papandreou G， Schroff F， et al. Rethinking atrous convolution for semantic image segmentation. 2017， DOI：10.48550/arXiv.1706.05587.
18	Chen L C， Zhu Y， Papandreou G， et al. Encoder-Decoder with atrous separable convolution for semantic image segmentation// European Conference on Computer Vision. Germany： Springer， 2018： 801-818.
19	Szegedy C， Liu W， Jia Y， et al. Going deeper with convolutions//Institute of Electrical and Electronic Engineers. 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston： Institute of Electrical and Electronic Engineers， 2015： 1-9.
20	Ronneberger O， Fischer P， Brox T. U-net： Convolutional networks for biomedical image segmentation//Medical Image Computing and Computer Assisted Intervention Society. Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015： 18th International Conference. Munich， Germany： Springer， 2015： 234-241.
21	He K， Zhang X， Ren S， et al. Deep residual learning for image recognition//Institute of Electrical and Electronic Engineers. 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas： Institute of Electrical and Electronic Engineers， 2016： 770-778.
22	Zhao H， Shi J， Qi X， et al. Pyramid scene parsing network//Institute of Electrical and Electronic Engineers. 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu： Institute of Electrical and Electronic Engineers， 2017： 2881-2890.
23	Sun K， Xiao B， Liu D， et al. Deep high-resolution representation learning for human pose estimation//Institute of Electrical and Electronic Engineers. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. California： Institute of Electrical and Electronic Engineers， 2019： 5693-5703.
24	Zhou X X， Wu Y L， Li M Y， et al. Automatic vegetation extraction method based on feature separation mechanism with deep learning. Journal of Geo-information Science， 2021， 23（9）： 1675-1689.
	周欣昕，吴艳兰，李梦雅，等. 基于特征分离机制的深度学习植被自动提取方法. 地球信息科学学报， 2021， 23（9）： 1675-1689.
25	Zhang Y， Wang H， Liu J， et al. A lightweight winter wheat planting area extraction model based on improved DeepLabv3+ and CBAM. Remote Sensing， 2023， 15（17）： 4156.
26	da Silva Mendes P A， Coimbra A P， de Almeida A T. Vegetation classification using DeepLabv3+ and YOLOv5//Institute of Electrical and Electronic Engineers. ICRA 2022 Workshop in Innovation in Forestry Robotics： Research and Industry Adoption. USA： Institute of Electrical and Electronic Engineers， 2022.
27	Hu Y N， An R， Ai Z T， et al. Researches on grass species fine identification based on UAV hyperspectral images in Three-River Source region. Remote Sensing Technology and Application， 2021， 36（4）： 926-935.
	胡宜娜，安如，艾泽天，等. 基于无人机高光谱影像的三江源草种精细识别研究. 遥感技术与应用， 2021， 36（4）： 926-935.
28	Zhang Y P， Wu X T， Li X L， et al. Identification of degraded grassland in Qinghai area of Yellow River Source based on high-resolution images. Acta Agriculturae Boreali-occidentalis Sinica， 2023， 32（2）： 198-211.
	张宇鹏，吴笑天，李希来，等. 基于高分影像的黄河源青海片区退化草地识别. 西北农业学报， 2023， 32（2）： 198-211.
29	Wen T， Liu X N， Ji T， et al. Studying on plant classification and recognition method for Three-River Source alpine grassland plant based on vegetation index. Acta Agrestia Sinica， 2022， 30（7）： 1811-1818.
	文铜，柳小妮，纪童，等. 基于植被指数的三江源高寒草地植物分类与识别方法研究. 草地学报， 2022， 30（7）： 1811-1818.
30	Lv C H， Liu Y Q. UAV-derived raster data of the Tibetan Plateau in 2020. National Tibetan Plateau/Third Pole Environment Data Center. https：//doi.org/10.11888/Geogra.tpdc.271124. https：//cstr.cn/18406.11.Geogra.tpdc.271124.
	吕昌河，刘亚群. 青藏高原无人机航拍栅格数据（2020）. 国家青藏高原数据中心. https：//doi.org/10.11888/Geogra.tpdc.271124. https：//cstr.cn/18406.11.Geogra.tpdc.271124.
31	Lv C H， Zhang Z M. UAV-derived raster data of the Tibetan Plateau （2021）. National Tibetan Plateau/Third Pole Environment Data Center. https：//doi.org/10.11888/Terre.tpdc.271903. https：//cstr.cn/18406.11.Terre.tpdc.271903.
	吕昌河，张泽民. 青藏高原无人机航拍栅格数据（2021）. 国家青藏高原数据中心. https：//doi.org/10.11888/Terre.tpdc.271903. https：//cstr.cn/18406.11.Terre.tpdc.271903.
32	Ouyang D， He S， Zhang G， et al. Efficient multi-scale attention module with cross-spatial learning// ICASSP 2023-2023 IEEE International Conference on Acoustics， Speech and Signal Processing （ICASSP）. Greece： IEEE， 2023： 1-5.
33	Fan Q H， Huang H B， Guan J Y， et al. Rethinking local perception in lightweight vision transformer. ArXiv， 2023， abs/2303.17803.

模型 Model	精确率 Precision （%）	召回率 Recall （%）	平均准确率 Mean precision （%）	每秒帧数 Frames per second
YOLOv5x	98.3	98.0	99.2	52
YOLOv5l	98.5	97.3	99.1	82
YOLOv5m	97.9	96.4	98.8	119
YOLOv5s	97.9	97.3	98.9	204

模型 Model	精确率 Precision （%）	召回率 Recall （%）	平均准确率 Mean precision （%）	每秒帧数 Frames per second
YOLOv5x	98.3	98.0	99.2	52
YOLOv5l	98.5	97.3	99.1	82
YOLOv5m	97.9	96.4	98.8	119
YOLOv5s	97.9	97.3	98.9	204

模型 Model	交并比 Intersection over union （%）	像素准确率Pixel accuracy （%）	参数量 Params	浮点运算次数 Giga floating-point operations per second
PSPnet	80.20	87.47	46.707	118.427
HRnet	82.27	87.39	29.538	79.915
Unet	84.55	89.78	43.933	184.100
DeeplabV3+	83.39	91.72	54.709	166.841
改进DeeplabV3+Improved DeeplabV3+	90.40	96.32	6.565	54.630

模型 Model	交并比 Intersection over union （%）	像素准确率Pixel accuracy （%）	参数量 Params	浮点运算次数 Giga floating-point operations per second
PSPnet	80.20	87.47	46.707	118.427
HRnet	82.27	87.39	29.538	79.915
Unet	84.55	89.78	43.933	184.100
DeeplabV3+	83.39	91.72	54.709	166.841
改进DeeplabV3+Improved DeeplabV3+	90.40	96.32	6.565	54.630

用地类型 Land use type	交并比 Intersection over union	像素准确率 Pixel accuracy	精确率 Precision	召回率 Recall
农业用地Agricultural land	88.96	96.48	90.54	96.48
工业用地Industrial land	87.74	95.57	91.80	95.57
生活用地Residential land	88.12	96.04	90.69	96.04
河流湿地Riverine wetlands	90.40	95.13	94.64	95.13
裸岩石区Bare rock area	93.12	95.05	97.64	95.05