Wetland Classification in Poyang Lake Using Dual-polarization Synthetic Aperture Radar Data with Feature Combination

Guanhua Zhou,Guifei Jing,Haoyu Miao,Zhiyuan Wang,Cheng Jiang

doi:10.18494/sam.2021.3590

Abstract

Wetland Classification in Poyang Lake Using Dual-polarization Synthetic Aperture Radar Data with Feature Combination

Highlights

The Poyang Lake wetland is one of the most important wetlands in the world, providing many ecological services, including purifying water, controlling natural hazards, and conserving soil and water, as well as numerous benefits for human society such as economic and scientific benefits.[1,2] Remote sensing techniques, mainly using optical, IR, and radar sensors, are widely used to map and monitor wetlands
Lönnqvist et al compared four methods of wetland classification using ALOS PolSAR full-polarization data, where two methods were based on supervised classification and two methods were based on unsupervised classification, and the intensity data were used.[11]. The results showed that the classification result of the full-polarization data was better than that of the intensity data regardless of the method used
Touzi proposed a new coherent target scattering vector model, which was applied to wetland classification and achieved good results.[12]. Patel et al used L-band and P-band full-polarization DLR-ESAR data to decompose a wetland target based on the feature vector, demonstrating the ability of PolSAR technology to characterize the different scattering performances of each component of a wetland ecosystem.[13]. Touzi proposed that the target scattering type can be represented by the symmetrical scattering type, and used the amplitude and phase for wetland classification, enabling the HH and VV phase difference and radiation scattering information to be distinguished; they used Touzi decomposition to classify wetlands and provide novel information.[14]. Liao and Wang used the

Summary

Introduction

The Poyang Lake wetland is one of the most important wetlands in the world, providing many ecological services, including purifying water, controlling natural hazards, and conserving soil and water, as well as numerous benefits for human society such as economic and scientific benefits.[1,2] Remote sensing techniques, mainly using optical, IR, and radar sensors, are widely used to map and monitor wetlands. Yamagata and Yasuoka conducted a wetland vegetation classification study using C-band ERS-1 SAR (VV polarization) and L-band JERS-l SAR (HH polarization) data with symbiotic matrix texture analysis.[5] Rio and Lozano-Garcia used RADARSAT-1 data processed by spatial filtering to study the classification of marsh and wetland, and achieved good precision, indicating the advantages of SAR in wetland classification.[6] Arzandeh and Wang applied gray-level co-occurrence matrix texture analysis to wetland classification using single-date RADARSAT images, and the results showed that this method can effectively improve the classification accuracy of wetlands.[7] Li and Chen proposed a rule-based wetland classification method using Landsat-7, RADARSAT-1, and digital elevation model (DEM) data, and the results showed that the confusion method of optical remote sensing and SAR was superior to the classification method using optical remote sensing alone or SAR alone.[8] A neural network combined with the Michigan Microwave Canopy Scattering (MIMICS) model, which was modified to fit to herbaceous wetland ecosystems, was used to retrieve information on wetland vegetation biomass in Poyang Lake with ENVISAT ASAR data.[9] Whitcomb et al used two seasons of L-band SAR imagery to produce a thematic map of wetlands throughout Alaska using the random forests decision tree (DT) algorithm and distinguished as many as nine different classes.[10] Full-polarization SAR systems are able to extract expanded information of the targets; recent studies have mainly focused on the applications of full-polarization SAR.

Objectives

Methods

Results

Conclusion