基于无人机与机器学习的黄河源高寒草地植被覆盖度反演技术

针对高原高寒地区大面积草地植被覆盖度调查与实验过程中地面测量效率低下,遥感数据质量不佳、数量源受限与反演结果不确定等问题,在黄河源地区利用低空无人机(unmanned aerial vehicle, UAV)遥感技术、统计建模与机器学习方法,开展基于可见光影像的高寒草地植被覆盖度反演与验证。结果表明:基于可见光构建的过绿指数与植被覆盖度的相关系数达0.676,比归一化差异指数高出近5.2%,具有较高的可靠性;利用过绿指数建立的高寒草地植被覆盖度统计模型中,对数模型和Gamma模型精度较高,但具有显著的地域差异性;直接利用低空无人机遥感波段值建立的机器学习模型精度显著优于各个统计模型,获得的均方根误差、估算精度、相对偏差和决定系数比统计模型中表现最优的对数模型分别提高2.68%、3.75%、7.35%和13.91%,且无需计算植被指数,在成本、效率和精度等方面具有较大的优势。

Vegetation Coverage Inversion of Alpine Grassland in the Source of the Yellow River Based on UAV and Machine Learning

In view of the low efficiency of ground measurement, poor quality of remote sensing data, limited quantity sources and uncertain inversion results during the investigation and experiment of large-scale grassland vegetation coverage in plateau alpine regions, the technology of low-altitude UAV, statistical modeling and machine learning methods were used to carry out inversion and verification of alpine grassland vegetation coverage based on visible light images in the source region of the Yellow River. The results of the study show that the correlation coefficient between the greening index and the vegetation coverage constructed based on visible light is 0.676, which is nearly 5.2% higher than the normalized difference index, and has high reliability; the alpine grassland vegetation coverage established by the greening index among the degree statistical models, the logarithmic model and the Gamma model have high accuracy, but they have significant regional differences; the accuracy of the machine learning model established directly using the band value of the low-altitude UAV is significantly better than that of each statistical model, and the root mean square error obtained , estimate accuracy, relative deviation and coefficient of determination are respectively 2.68%, 3.75%, 7.35% and 13.91% higher than the logarithmic model with the best performance in the statistical model, and there is no need to calculate the vegetation index. It has great advantages in terms of cost, efficiency and accuracy.