Deep Learning Based Extraction And Spatial-Temporal Evolution Of Lake Wetland

Lake wetland is an ecosystem with multiple ecology function,which is very important for the sustainable development of human and ecological environment.As environmental degradation worsens,the wetland area is decreasing.We should protect and recover wetlands immediately.To provide important data support for wetland protection and decision making,we could use remote sensing technology to extract the spatial distribution information of wetlands,and monitor the temporal and spatial evolution of lake wetland dynamically.Spatial and spectral information are often not considered simultaneously in extracting remote sensing wetland with commonly used methods,so it is very difficult to ensure the reliability and accuracy of wetland extraction.In this paper,we use a deep learning algorithm during the using of remote sensing spectral characteristics,and construct the joint probability for wetland extracting.We could analysis the temporal and spatial variation according to the different periods of lake wetland information.This paper mainly include:(1)The construction of wetland recognition model based on deep learning.The method of deep learning is introduced to construct multi-scale convolution neural network model for wetland identification.And select the wetland sample to train the wetland recognition model based on deep convolution neural network.(2)Study of the joint estimation model with spectral-spatial approach.Based on the deep learning convolution neural network model,combined with the normalized differential water body index and the humidity component,the wetland is extracted by considering the spatial characteristics and spectral characteristics of the wetland.(3)The comparison between the current wetland extraction methods and the approach based on deep learning.Do contrast experiment with the minimum distance method,maximum likelihood method,decision tree classification method,object-oriented method,support vector machine method and extraction method based on deep learning.Then,evaluating the accuracy and testing the feasibility of the method based on deep learning.(4)Research of the temporal and spatial evolution among the lake wetland resources.This paper takes Poyang Lake as an example to analyze the temporal and spatial variation of lake wetland.By using the landscape ecology method to analyze the wetland information of the lake in 2000,2005,2010 and 2015,we could know the trend of wetland change process with the landscape pattern index and put forward the corresponding suggestion for protection.