多源遥感数据与机器学习算法的沼泽湿地信息提取

Extraction of Swamp Wetland Information Based on Multi-source Remote Sensing Data and Machine Learning Algorithm

基于多源遥感数据与机器学习算法,旨在探讨提高沼泽湿地信息提取精度的方法。利用GEE(Google Earth Engine)平台获取的Sentinel-1合成孔径雷达(SAR)数据、Sentinel-2多光谱数据以及地形数据,结合多种机器学习算法提取沼泽湿地遥感信息,并在大兴安岭山区的南瓮河自然保护区以及松嫩平原区的莫莫格保护区进行精度验证。结果表明:利用多源遥感数据结合机器学习算法的研究方法,可显著提高沼泽湿地信息提取精度。在平原地区沼泽湿地信息提取中,雷达数据比多光谱数据贡献度更高;在山地地区沼泽湿地信息提取中,二者具有相近的贡献度。Sentinel-1、Sentinel-2数据和地形湿度指数数据的组合更有利于各种地形沼泽湿地信息提取。随机森林机器学习算法对沼泽湿地信息提取精度最高。

This article is based on multi-source remote sensing data and machine learning algorithms,aiming to explore methods to improve the accuracy of swamp wetland information extraction.By using the Sentinel-1 synthetic aperture radar(SAR)data,Sentinel-2 multi-spectral data and terrain data obtained from the GEE(Google Earth Engine)platform,combined with a variety of machine learning algorithms,the remote sensing information of swamp wetlands was extracted and analyzed.Verification tests were carried out in the Nanwenghe nature reserve in the Daxingan mountains and the Momoge nature reserve in the Songnen plain area.The results show that the research method using multi-source remote sensing data combined with machine learning algorithms significantly improve the accuracy of swamp wetland information extraction.In the extraction of swamp wetland information in plain areas,radar data has a higher contribution than multispectral data;In the extraction of swamp wetland information in mountainous areas,the two have close contributions.The combination of Sentinel-1,Sentinel-2 data and terrain humidity index data is more conducive to the extraction of swamp wetland information in various terrains.The random forest is the most effective machine learning algorithm for extracting swamp wetland information,achieving the highest accuracy.