| Mangrove refers to the woody community of tidal flat wetland that grows in intertidal zone of tropical and subtropical low-energy coast and is composed of evergreen shrubs or trees mainly composed of mangrove plants.Mangrove wetland is an important ecological wetland and species habitat,which has the ecological function and ecological service value of protecting the shore from waves,regulating the climate and maintaining the coastal diversity.Due to its long time and heavy workload,it is difficult to accurately locate its boundary due to its growth in intertidal shoals and silt.Satellite remote sensing image has the advantages of wide coverage,high aging and unique periodicity,and can effectively obtain mangrove information.Therefore,monitoring its distribution by remote sensing technology is of great significance for the protection and management of mangrove wetlands.In this study,Sentinel-1 and Sentinel-2 data from 2018 were used to extract mangrove information from Maoweihai Bay of Guangxi,so as to provide important basis for monitoring,protection and planning of mangrove ecosystem.Firstly,the image is segmented by the most extensive classification network evolutionary algorithm,and features of image objects are extracted based on object oriented technology.Secondly,the random forest algorithm is used for feature optimization,and on this basis,a variety of different feature combination schemes are constructed to perform classification,so as to explore the effectiveness of the feature selection method.Finally,object-oriented KNN,SVM and RF method were used to identify the object types in the studied area,and used confusion matrix to analyze the classification results of the three methods.The following conclusions were drawn:(1)Based on the fused single time Sentinel-1 and Sentinel-2 data,the image was segmented and feature extracted by object-oriented techniques.The results show that the segmentation scale and parameter sizes of images could be obtained effectively with the help of multi-scale segmentation evaluation tools and the method of artificial aided determination.(2)Among the results of the feature combination classification of each group,the classification results of the multi-feature coupling optimization model obtained by the random forest algorithm obtained good extraction accuracy,with the overall accuracy of 89.60% and Kappa coefficient of 0.8756.It showed that the algorithm can retain the features useful for classification,reduced the redundancy and improved the classification efficiency.(3)This paper adopted the object-oriented KNN,SVM and RF method to identify the image feature types respectively,and used the confusion matrix to evaluate the recognition results.The results showed that the overall accuracy and Kappa of the object-oriented RF method were the highest,which was 3.4% and 0.0434 higher than the object-oriented KNN method,and 0.83% and 0.0111 higher than the object-oriented SVM method,indicated that the random forest algorithm had a stable classification performance.(4)Based on the fusion of the Sentinel data,this paper used the random forest algorithm to extract the mangrove information of Maoweihai Bay in Guangxi.The mapping accuracy and user accuracy of identification were 96.39% and 97.56%,respectively.The area was 19.2km,accounted for 2.67% of the whole research area.This study revealed the potential of Sentinel-1 and Sentinel-2 data in mangrove monitoring,which could provide important basis for monitoring,conservation and planning of mangrove ecosystems.The purpose of this study is to explore the application value of Sentinel-1 radar image and Sentinel-2 optical image in mangrove monitoring,and to provide a technical reference for mangrove remote sensing identification and monitoring,so as to provide a scientific reference for the protection of mangrove wetland ecosystem. |
PDF Full Text Request