基于无人机影像的高寒草甸退化斑块识别与应用OA

Patchily degradation of alpine meadows was an important feature of alpine grassland degradation on the Qinghai-Tibetan Plateau,and the identification of patchily degraded alpine meadows through UAV imagery could accurately reflect the degradation of alpine meadows at large scales,which was of great significance for the protection and restoration of alpine meadows.In this study,patchily degraded alpine meadows areas were selected in typical catchments of the Qinghai Lake region and the Yellow River source area to collect imagery and high-precision topographic data using UAVs.Different feature selection schemes were designed based on importance ranking and correlation analysis separately.Object-oriented UAV imagery classification was performed using various machine learning classifiers to achieve multi-feature refined automatic identification of degraded alpine meadow patches,and healthy meadow,revegetated patch,bare patch,mound and rat hole was identified.Correlation analysis was then conducted between the identification results and topographic data to explore large-scale survey methods for patchily degraded alpine meadows based on UAVs and their application potential.The results showed that:(1)UAV imagery combined with object-oriented classification was highly suitable for identifying degraded alpine meadow patches,with an overall accuracy exceeding 96％,and different classifiers are suitable for the identification of different patches.The order of identification accuracy from highest to lowest is as follows:rat hole,mound,healthy meadow,bare patch and revegetated patch.(2)Feature selection based on importance ranking is more suitable for identifying degraded alpine meadow patches than that based on correlation analysis.Spectral and texture features were more important than geometric features in patch identification,and the Bayes classifier performed best.(3)The degradation and revegetation of alpine meadow showed significant correlations with elevation,slope,and curvature,but no obvious correlation with rodent damage.Healthy meadows and restored patches show a significant negative correlation with elevation(P<0.05),and an extremely significant negative correlation with slope and curvature(P<0.01);bare patches show a significant positive correlation with elevation,slope and curvature;all types of patches have no significant correlation with the density of rodent holes(P>0.05).It is further found that the dominant factors of alpine meadow degradation exhibited spatial scale differentiation.This study proposed a UAV data-based,object-oriented refined identification and terrain correlation analysis method,which provided a novel technical pathway for precisely formulating alpine meadow restoration measures and evaluating rehabilitation effectiveness.