Classification Of Gangenore Lake Wetland Coverage Based On Multi-frequency Multi-polarization SAR Data

Polarimetric SAR image classification is a very important research issue in radar remote sensing applications.With the introduction of polarization scattering features and nonparametric model classification methods,the research on polarization SAR target decomposition theory and machine learning algorithms has been further developed,but due to the high dimensional characteristics of polarized SAR images,it is used for classification models.Feature set information redundancy.Aiming at this problem,this paper studies the characteristics of full-polarization SAR and dual-polarization SAR data,and studies the classification of surface coverage based on polarization scattering feature and support vector machine(SVM)algorithm.This study selected Gangenore Lake and its surrounding area of 315 square kilometers in the Dalinor Wetland Reserve of Keshiketeng Banner,Chifeng City,Inner Mongolia Autonomous Region as the study area,and obtained Radarsat-2 full polarization data and the same area.The ALOS-2 dualpolarized image first analyzes the difference between the two bands for ground recognition.Then,the characteristics of the full polarization different polarization decomposition method and the data after fusion with the optical image are analyzed.Finally,the SVM classifier is used.The surface cover classification,and the classification accuracy test is performed based on the high-resolution interpretation map and the Google Earth image.The two images are classified and compared,and the influence of input features and algorithms on the classification is compared and analyzed.(1)ALOS-2 L-band has a longer wavelength and good penetrability,and has better recognition of terrestrial vegetation with high vegetation coverage and obvious vertical structure difference,and has good recognition effect on bare land;Radarsat-2 C-band wavelength is relatively high.Short,single identification from each polarization channel is likely to cause confusion between the land types.(2)Among the features of the polarization analysis of the fully polarized image,Pauli decomposition has better recognition effect on vegetation-covered areas such as bare land and water bodies,as well as forest land and grass marshes;H/A/ Decomposition in forest land,bare land,rural settlements,and water bodies is better;Freeman decomposition,NNED decomposition,and Yamaguchi decomposition are suitable for identifying the main scattering mechanisms,such as bare land,woodland,and water surface.,cultivated land,rural settlements,swamps,etc.(3)The color difference between n the dual-polarized SAR image and the optical image is obvious,and the recognition effect on bare land,cultivated land,rural residential area,saline-alkali land,etc.is better;the ground separation of fully polarized SAR image and optical data Good degree,can clearly identify bare land,vegetation cover area,sandy land and saline-alkali land.(4)Regardless of the SVM classification of the fully polarized SAR image feature set of the fused data,the overall classification accuracy is 75%,and the SVM classification accuracy considering the fusion data is 85%;the dualpolarization SAR image feature set of the fusion data is not considered.SVM classification,the overall accuracy of classification is 39%,and the SVM classification accuracy considering fusion data reaches 81%.It can be seen that the fusion data has a positive effect on SVM classification.(5)Full-polarization data feature set The SVM classification of the feature selection algorithm has an overall accuracy of 84%,which is not much different from the full feature set classification accuracy,but the feature set is compressed by 45%,the number of support vectors is reduced,and the computational efficiency is greatly improved.The dual-polarized data feature set uses the feature selection algorithm to optimize the overall accuracy of the SVM classification by 83%,and the feature number in the feature set is reduced from 38 to 20,which has a positive effect on the classification accuracy and efficiency improvement;From the selection results,it is confirmed again that the fusion data is very helpful for the improvement of classification accuracy.