基于“网格搜索+XGBoost”算法的浅海遥感水深反演

Inversion of Remote Sensing Water Depth in Shallow Water Based on “Grid Search+XGBoost” Algorithm

传统船载水深测量受船只吃水影响,难以在浅水区域开展,遥感水深反演作为传统方法的有益补充,其重要性日益凸显。以GF-1多光谱影像为数据源,以船载声呐实测水深点作为训练样本和检测样本,以相关系数、均方根误差和平均绝对误差作为评价指标,首次将网格搜索+XGBoost模型应用于启东恒大威尼斯浅海区域水深反演。实验表明,网格搜索+XGBoost模型水深反演的相关性系数达到0.820,均方根误差0.247 m,平均绝对误差0.134 m,与GBDT模型和波段比值模型相比,其水深反演精度更高,且易于实现。该研究方法和成果为快速获取大范围浅海水深提供了借鉴,为相关水上勘察和海洋资源开发提供了技术思路。

The traditional ship borne sounding is difficult to carry out in shallow water because of the ship’s draft. As a useful supplement to the traditional bathymetric survey,the importance of the remote sensing water depth inversion becomes increasingly visible. In this paper,the GF-1 multispectral image is used as the data source,the measured depth points of ship borne sonar are used as the training samples and detection samples,and the correlation coefficient,root mean square error and mean absolute error are used as the evaluation indexes. The grid search+XGBoost model is applied to the remote sensing water depth inversion in the shallow sea of Qidong for the first time. In the experiment,the correlation coefficient of the grid search+XGBoost model is 0.820,the root mean square error is 0.247 m,and the average absolute error is 0.134 m. Compared with GBDT model and Band Ratio model,the accuracy of depth inversion is higher and easy to implement. The research methods and results of this paper provide a reference for the rapid acquisition of large-scale shallow water depth,and provide technical ideas for related marine investigation and resources development.