Font Size: a A A

Remote Sensing Dynamic Monitoring Of Coastal Wetland Vegetation Based On Fully Convolutional Networks Combined With Object-Oriented Method

Posted on:2020-07-01Degree:MasterType:Thesis
Country:ChinaCandidate:J Y XieFull Text:PDF
GTID:2370330578961349Subject:Forest management
Abstract/Summary:PDF Full Text Request
Coastal wetlands play an important role in ecological protection,environmental regulation,biodiversity conservation,resource utilization,disaster prevention,they are also important prerequisites for ensuring regional sustainable development.Wetland vegetation is an important part of the coastal wetland ecosystem,has an indicative role for the coastal wetland ecological environment.Studying the succession of coastal wetland vegetation is of great significance for understanding the evolution mechanism and trend of coastal wetlands.Accurately monitoring the spatial distribution and dynamic changes of coastal wetland vegetation is an important basis for studying coastal vegetation.This paper taked part of the coastal wetland in the south bank of Hangzhou Bay as the research area,the three-phase high spatial resolution remote sensing data and field survey data were used to monitor the temporal and spatial dynamics of typical vegetation in coastal wetlands.Firstly,the research explored the remote sensing classification method of wetland vegetation based on the 2013 QuickBird image and field survey data.The images were processed by three methods: fully convolutional networks(FCN),random forest combined with object-oriented method and FCN combined with object-oriented method.The accuracy evaluation results showed that the FCN combined with object-oriented method is better.Then,this method was used to classify the QuickBird image in 2008 and the GF-2 image in 2017 to extracte the coastal vegetation distribution map,and finally analyzed the temporal and spatial changes of the typical vegetation of coastal wetland in the study area.The influencing factors caused the change were explored to provide a basis for the rational development and utilization of coastal wetlands.The main conclusions of this paper are as follows:(1)The FCN which is more accurate than the random forest could extract the typical vegetation of coastal wetland fine,but it was easy to produce salt and pepper phenomenon.Combining with object-oriented could remove this phenomenon and effectively improve the classification accuracy of wetland vegetation.(2)The effect of object-oriented segmentation was greatly affected by the segmentation scale,shape parameters and compactness index.In this paper,the average global scoring index was used to quantitatively evaluate these parameters.Finally,the optimal segmentation scale,the shape weight and the compactness index weight was 160,0.2 and 0.2 respectively.(3)The typical vegetation of coastal wetlands in the study area mainly includeed Scirpus mariqueter,Spartina alterniflora and Phragmites communis.The spatial distribution characteristics were following: the Scirpus mariqueter was mainly distributed in the intertidal zone near the coast,that was,the new tidal flat.It played an important role in transforming the ecological environment of the tidal flat.The Spartina alterniflora crossed with the Scirpus mariqueter,mainly distributing in the silt area of the coastal shoal.Because it was an invasive specie,it had been manually removed in recent years,which caused it decreasing.The Phragmites communis was distributed in the aquaculture pond at the beginning of the transformation,and the area is less while getting inland,and artificial vegetation was dominanted.Driven by the natural evolution of coastal zones and human activities,the ecological environment of coastal wetlands varied greatly in time and space,which caused the spatial distribution of typical vegetation in wetlands.(4)The land use of Hangzhou Bay changed drastically.The human activities such as cofferdam,urbanization and aquaculture pond expansion had a great impact on the typical vegetation of coastal wetlands.The aquaculture ponds in the study area increased significantly between 2008 and 2017,and the area of tidal flats continued to decrease,resulting in a significant reduction in the area of the Scirpus mariqueter.At the same time,the destruction of vegetation habitats,the artificial control and treatment of Spartina alterniflora made it extinct,and the adaptability of the Phragmites communis and its certain ornamental nature made it a dominantion.Especially after the shoal cofferdam was transformed into a culture pond,it provided growth conditions for the Phragmites communis.It is concluded that the influencing factors of spatial and temporal changes of typical vegetation in coastal wetlands are mainly human activities,supplemented by natural evolution.The ecological environment of wetland vegetation is seriously disturbed by humans.In order to protect the wetland,the vegetation should be naturally grown and succeeded,and the development of tidal flats should be slowed down,and the artificial activities of the coastal zone should be reduced,and the development of tidal flats,and the growth of coastal wetland vegetation should maintain a moderate balance.
Keywords/Search Tags:FCN, object-oriented, typical vegetation in coastal wetlands, dynamic monitoring, high resolution
PDF Full Text Request
Related items