Scholarlink

<p>Desert vegetation can effectively curb the process of desertification and protect the ecological security of oases. Based on the characteristics of high resolution and convenience of Unmanned Aerial Vehicle (UAV) images, it has become possible to quickly obtain the vegetation coverage of oases in the desert hinterland. In this study, the natural vegetation of Dariyabui Oasis in the desert hinterland was taken as the object of study, and multiple types of DJI UAVs, such as DJI Phantom 4 (Pro, Multispectral, RTK) were used to acquire images of typical vegetation samples in the summer and autumn growing seasons. The Random Forest (RF), Support Vector Machine (SVM), Extreme Learning Machine (ELM), and eXtreme Gradient Boosting (XGboost) machine learning regression models were constructed with OTSU, Excess green-Excess red (ExGR), and RF extraction Fractional Vegetation Cover (FVC) as independent variables and visual interpretation FVC as dependent variable, respectively. The best regression model was screened by R<sup>2</sup>, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) evaluation metrics. The independent variables (OTSU, ExGR, RF extraction FVC) of the best regression models were ranked in order of importance, and the optimal extraction method was determined by the degree of contribution to the model prediction results. The results showed that the RF regression model was the best model for predicting FVC during the summer (R<sup>2</sup> = 0.95, RMSE = 0.22%, MAE = 0.13%) and autumn (R<sup>2</sup> = 0.98, RMSE = 0.14%, MAE = 0.08%) growing seasons in the desert hinterland oasis. The integration of the OTSU algorithm and Image J interaction is able to batch and accurately extract desert oasis vegetation sample FVC.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative Machine Learning Algorithms for Dynamic Vegetation Cover Retrieval in Desert Oasis from UAV Time-Series Data

  • 期刊论文

  • 发表日期：

    2025年9月10日

查看全文
  • Ning Wang,
  • Fuxin Jin,
  • Chaoyong Peng,
  • Peng Wang,
  • Yuchuan Guo,
  • Ce Yu

摘要

Desert vegetation can effectively curb the process of desertification and protect the ecological security of oases. Based on the characteristics of high resolution and convenience of Unmanned Aerial Vehicle (UAV) images, it has become possible to quickly obtain the vegetation coverage of oases in the desert hinterland. In this study, the natural vegetation of Dariyabui Oasis in the desert hinterland was taken as the object of study, and multiple types of DJI UAVs, such as DJI Phantom 4 (Pro, Multispectral, RTK) were used to acquire images of typical vegetation samples in the summer and autumn growing seasons. The Random Forest (RF), Support Vector Machine (SVM), Extreme Learning Machine (ELM), and eXtreme Gradient Boosting (XGboost) machine learning regression models were constructed with OTSU, Excess green-Excess red (ExGR), and RF extraction Fractional Vegetation Cover (FVC) as independent variables and visual interpretation FVC as dependent variable, respectively. The best regression model was screened by R2, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) evaluation metrics. The independent variables (OTSU, ExGR, RF extraction FVC) of the best regression models were ranked in order of importance, and the optimal extraction method was determined by the degree of contribution to the model prediction results. The results showed that the RF regression model was the best model for predicting FVC during the summer (R2 = 0.95, RMSE = 0.22%, MAE = 0.13%) and autumn (R2 = 0.98, RMSE = 0.14%, MAE = 0.08%) growing seasons in the desert hinterland oasis. The integration of the OTSU algorithm and Image J interaction is able to batch and accurately extract desert oasis vegetation sample FVC.