基于无人机和机器学习的川西北修复沙地植被信息提取  

Vegetation Information Extraction for Restoration of Sandy Land in Northwest Sichuan Based on Unmanned Aerial Vehicles and Machine Learning

作　　者：徐渝杰 舒向阳 陶敏 孙奕函 刘唯佳 董高成 何沁 李杰 李一丁 邓良基 杨雨山 XU Yujie;SHU Xiangyang;TAO Min;SUN Yihan;LIU Weijia;DONG Gaocheng;He Qin;LI Jie;LI Yiding;DENG Liangji;YANG Yushan(College of Resource,Sichuan Agricultural University,Chengdu 611130,China;Key Laboratory of Land Resources Evaluation and Monitoring in Southwest,Ministry of Education,Sichuan Normal University,Chengdu 610068,China;The Institute of Geography and Resources Science,Sichuan Normal University,Chengdu 610068,China;Chengdu Academic of Agriculture and Forestry Sciences,Chengdu 611130,China;Renshou Shujin Real Estate Co.,Ltd,Meishan 620500,Sichuan,China)

机构地区：[1]四川农业大学资源学院,成都611130 [2]四川师范大学西南土地资源评测与监测教育部重点实验室,成都610068 [3]四川师范大学地理与资源科学学院,成都610068 [4]成都市农林科学院,成都611130 [5]仁寿蜀锦置业有限公司,四川眉山620500

出　　处：《四川农业大学学报》2024年第1期181-187,共7页Journal of Sichuan Agricultural University

基　　金：四川省科技计划项目(2021JDRC0082)。

摘　　要：【目的】旨在无人机影像中提取植被信息(草本和灌木),估算植被覆盖度,用于生态修复领域反映植被长势和丰度。【方法】选取水体、灌木、草本和沙地4类地物,采用4种机器学习算法,深度学习、马氏距离、最大似然法和最小距离法进行精度对比,选取精度最高的算法作为研究方法。【结果】4种方法得到总体精度分别为95.47%、95.14%、93.30%和71.98%,kappa系数分别为0.92、0.91、0.88和0.57。【结论】基于深度学习方法分析发现,红原沙化治理示范基地范围内灌木、草地、水体和沙地面积分别为0.09、0.14、0.04和0.32 km^(2)。该方法可以为川西北高寒修复沙地监测、研究与治理状况评价提供数据支持和一定的科学依据。【Objective】This study aims to extract vegetation information(herbs and shrubs)from UAV images,and estimate vegetation coverage,finally reflecting vegetation growth and abundance in the field of ecological restoration.【Method】Four types of surface objects including water,shrubs,herbs and sand were selected,and four machine learning algorithms,including deep learning,Mahalanobis distance,maximum likelihood method and minimum distance method,were used for precision comparison.The algorithm with the highest accuracy is selected as the research method.【Result】The overall accuracy of the four methods were 95.47%,95.14%,93.30%and 71.98%,and kappa coefficient were 0.92,0.91,0.88 and 0.57,respectively.【Conclusion】The optimal method of the four algorithms was the deep learning method.The water body and sandy land are 0.09,0.14,0.04 and 0.32 km^(2),respectively.This method can provide data support and scientific basis for monitoring,research and management evaluation of alpine restoration sandy land in northwest Sichuan.

关 键 词：高分辨率无人机影像 沙地植被信息提取 植物覆盖率 机器学习 

分 类 号：S812[农业科学—草业科学]