一种基于机器学习综合指数的遥感赤潮识别方法
A remote sensing red tide identification method based on machine learning composite index
李颖 1秦勉 1谢铭 1王作敏1
作者信息
- 1. 大连海事大学航海学院,辽宁大连 116026
- 折叠
摘要
基于5种可以指示赤潮爆发的子指数及其综合方式,提出一种可用于多光谱卫星图像中赤潮识别的线性综合识别指数(LCI)方法.根据大连市星海湾现场实测的多日赤潮连续监测光谱数据,开展支持向量机与逻辑回归分类,并得到分类平面系数作为归一化的子指数组合权重,获得了两种较佳的赤潮识别指数组合方式,该两种方式在验证集上均取得了 F1分数0.86以上的分类效果,且LCI数值与赤潮发生的概率也呈正相关.将得到的分类系数应用于多光谱卫星图像的赤潮像元提取,结果显示,两种线性综合指数均在图像中成功提取出赤潮区域.本文提出的赤潮综合识别指数为遥感图像中的赤潮识别提供了新的光谱指数运算方法.
Abstract
A linear comprehensive identification index(LCI)method was proposed for red tide identification in multispec-tral satellite images based on five sub-indexes and their com-prehensive methods which can indicate red tide outbreaks.Based on the multi-day continuous monitoring spectral data of red tide in Xinghai Bay,Dalian City,the support vector ma-chine and logistic regression classification were carried out,and the classification plane coefficients were obtained as nor-malized sub-index combination weights.Two optimal combina-tions of red tide identification indexes were obtained,both of which achieved a classification performance of F1 score above 0.86 on the validation set,and the LCI value was positively correlated with the probability of red tide occurrence.Further-more,the obtained classification coefficients were applied to the extraction of red tide pixels in multispectral satellite ima-ges,whereby the results show that both linear composite inde-xes have successfully extracted red tide areas in the images.The comprehensive identification index of red tide proposed in this paper provides a new spectral index calculation method for red tide identification in remote sensing images.
关键词
赤潮识别/多光谱遥感图像/实测光谱数据/机器学习分类/线性综合识别指数(LCI)Key words
red tide identification/multispectral remote sens-ing image/actual measured spectral data/machine learning classification/linear comprehensive identification index(LCI)引用本文复制引用
基金项目
辽宁省"兴辽英才计划"项目(XLYC2001002)
大连市高层次人才团队创新支持计划(2022RG02)
中央高校基本科研业务费专项资金资助项目(3132023507)
出版年
2024
NETL
NSTL
万方数据