基于模糊隶属度的多核SVR遥感水深融合探测

遥感是海岸带浅海和岛礁周边水深探测的重要手段，支持向量回归(SVR)是广泛应用于数据回归的机器学习模型。本文将SVR引入多光谱遥感水深探测，针对单核SVR模型在浅水区水深反演中误差较大的问题，以单核SVR模型反演不同水深段的模糊隶属度作为决策融合因子，提出了基于模糊隶属度的多核SVR遥感水深融合探测方法，并以我国西沙群岛中的北岛为实验区，与单核SVR模型和传统的对数线性水深遥感模型开展对比实验。实验结果表明:1)基于模糊隶属度的多核SVR融合模型在25 m以浅的水域，平均绝对误差0.99 m，平均相对误差8.2%；2)融合模型的平均相对误差分别比以RBF、Sigmoid、多项式、线性为核函数的四种单核SVR模型提高了1.7%、4%、4.4%、4.8%，比对数线性模型提高了5.5%；3)对于不同水深段，多核SVR融合模型在0~15 m的3个水深段内平均相对误差比四种单核SVR模型提高了0.7%至54.9%不等，在0~25 m的5个水深段内比对数线性模型提高了1.1%至20.4%不等。

Multiple kernel support vector regression based on fuzzy membership for remote sensing water depth fusion detection

Remote sensing is an important means of water depth detection in coastal shallow waters and reefs. Support vector regression (SVR) is a machine learning model which is widely used in data regression. In this paper, SVR is introduced into multi-spectral remote sensing water depth detection. Aiming at the problem that the single-kernel SVR model has a large error in shallow water depth inversion, the fuzzy membership degree of different water depth is retrieved as a decision fusion factor with single kernel SVR model, a multi kernel SVR remote sensing water depth fusion detection based on fuzzy membership degree is put forward. And taking the North Island of the Xisha Islands in China as an experimentation area, the comparison experiments with the single kernel SVR models and the traditional log-linear bathymetric remote sensing model are carried out. The results show that:1) In range of 0 to 25 meters, the mean absolute error(MAE)of the multi kernel SVR remote sensing water depth is 0.99 m, the mean relative error(MRE) is 8.2%; 2) Compared to the RBF-SVR model、Sigmoid-SVR model、Poly-SVR model and Line-SVR model, the MRE of the fusion detection model improved 1.7%、4%、4.4%、4.8%, compared to traditional log-linear model, the MRE improved 5.5%; 3) For different water depth section, the 3 sections in 0~15 m depth, compared to 4 different single kernel SVR models, the MRE of fusion detection model improved 0.7% to 54.9%, in the 5 sections of 0~25 m, compared to traditional log-linear model, the MRE of fusion detection model improved 1.1% to 20.4%.