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

【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.