基于对象级特征与XGBoost的城市河流提取方法

Urban River Extraction Method Based on Object-Level Features and XGBoost Algorithm

摘要: 为准确提取城市河流信息，本文提出一种基于对象级特征与XGBoost的城市河流提取方法：首先利用NDWI指数对GF-1号影像进行像素级的水体提取；接着将提取结果形成水体对象，并计算对象形状特征，构建样本集；然后对样本进行打标，并利用样本集训练XGBoost算法，通过调整超参数，使XGBoost算法适应实际样本情况；最后，利用训练好的XGBoost算法进行河流对象提取，并进行提取精度分析。实验证明，本方法能减少城市阴影的干扰，并有效区分河流与湖泊，河流提取的准确率达到90.91%。

Abstract: To extract urban river information accurately, this paper proposes an object-level feature and XGBoost-based method for urban river extraction. The methodology comprises the following steps: first, the NDWI index is applied to GF-1 imagery for pixel-level water extraction; subsequently, the extraction results are formed into water body objects, with object shape features calculated to construct a sample set; then, the samples are labeled and used to train the XGBoost algorithm, where hyperparameter tuning is performed to adapt the algorithm to the actual sample characteristics; finally, the trained XGBoost algorithm is employed for river object extraction. Experimental results demonstrate that this method effectively minimizes interference from urban shadows, and successfully distinguishes rivers from lakes, achieving a river extraction accuracy of 90.91%.