顾及空间自相关特征的机器学习水深反演方法研究

基于多光谱影像的水深反演方法是获取近岸水深信息的高效手段，然而反演精度低一直是制约其广泛应用的瓶颈。本文聚焦于实测水深与多光谱数据自身的空间自相关特性，提出在机器学习框架下将学习样本的空间自相关特征与统计互相关特征相结合，以提高水深反演精度。西沙北岛海域的实验结果表明:在实测数据量较小的情况下，相比传统机器学习，顾及自相关特征的新方法可获得18%的精度提升；而当实测数据量充足时，精度提升可达到27%。结果表明，将数据源的空间自相关特征融入机器学习算法中，可显著提升多光谱水深反演结果的精确性，进而为浅海海洋研究提供有效数据支撑。

Retrieving shallow bathymetry by integrating spatial autocorrelation features with machine learning

Retrieving shallow water depth based on multispectral satellite imagery is highly cost-effective. However, the extensive application of satellite-derived bathymetry has been restricted by its low prediction accuracy. To improve about the accuracy of the retrieved bathymetry, spatial autocorrelation features within the in situ depth measurements and the multi-spectral image are focused in this research. To this end, we develop a machine learning method combining with spatial autocorrelation features and statistical intercorrelation features of learned samples. The experimental results of Xisha Beidao show that compared with the traditional machine learning, the accuracy of the new method is improved by 18% when the number of in situ depths is small. On the contrary, when the number of in situ depths is large, an improvement of 27% in root mean square error is achieved. This demonstrates that incorporating the spatial autocorrelation features of data sources into the machine learning can significantly improve the prediction accuracy, and then provide effective data support for shallow ocean research.