一种融合多波段比值法和随机森林算法的Sentinel-2绿潮信息提取方法
A Sentinel-2 green tide information extraction method incorporating multi-band ratio method and Random Forest algorithm
周红武 1宁珊 1李达 1潘锡山 2李乔 1赵敏 1唐潇1
作者信息
- 1. 南京师范大学 海洋科学与工程学院,江苏 南京 210023
- 2. 江苏省海涂研究中心,江苏 南京 210023
- 折叠
摘要
遥感卫星可以高效、大面积地监测海洋绿潮的发生及发展情况,而绿潮的精准识别提取尤为重要,对于维护海洋生态环境安全具有重大意义.本研究提出一种融合多波段比值法和随机森林算法(random forest,RF)的绿潮信息提取方法,基于高分辨率的Sentinel-2遥感影像,以苏北近岸海域为研究区,以绿潮和非绿潮两类地物为识别目标进行精准识别分类,并与最大似然法(maximum likelihood classification,MLC)、RF和支持向量机(support vector machine,SVM)以及融合多波段比值法的MLC和 SVM等五种模型进行对比.结果表明:归一化植被指数(normalized difference vegetation index,NDVI)和漂浮藻类指数(floating algae index,FAI)能够有效提高绿潮信息提取的准确性,融合这两种指数的RF模型分类结果Kappa系数为 0.98,总体精度为 99.92%,高于其他五种模型.研究结果可为海洋环境生态及生物地球化学循环的相关研究与治理提供理论与技术支撑.
Abstract
Remote sensing satellites enable an efficient and extensive monitoring of the occurrence and development of marine green tides,while the accurate identification and extraction of green tides are significantly important for the protection of the marine ecological environment.The present study proposed a green tide information extraction method with the combinations of the multi-band ratio method and RF algorithm.Based on the high-resolution Sentinel-2 imagery,we distinguished between the green tide and non-green tide features in northern Jiangsu province offshore area.Five other models were compared with our developed identification avenue,including MLC,RF,SVM,MLC with multi-band ratio method,and SVM with multi-band ratio method.The results showed that the two vegetation indices,NDVI and FAI could markedly improve the accuracy of green tide information extraction.The RF model,which combined these two vegetation indices,provided a Kappa coefficient of 0.98 and an overall accuracy of 99.92%,outperforming the other five models.These findings of our study could offer,to some extent,the theoretical and operational supports for the future related research and the scientific management of the ecological and biogeochemical processes in marine environments.
关键词
黄海绿潮/Sentinel-2/NDVI/FAI/随机森林Key words
green tide/Sentinel-2/NDVI/FAI/random forest引用本文复制引用
基金项目
江苏省海洋科技创新专项项目(JSZRHYKJ202206)
江苏省生态环境监测科研基金项目(2329)
出版年
2024
NETL
NSTL
维普
万方数据