Research On Water Information Extraction Method Of Multi-source Remote Sensing Based On Deep Learning And Its Application

Surface water is an important component of the earth's water resources.Rapid and effective monitoring of surface water plays a vital role in the earth's ecological balance and stability.Compared with traditional on-site measurement methods,remote sensing technology has a wide monitoring scale and high timeliness,and is widely used in surface water extraction and dynamic evolution monitoring.At present,remote sensing water body extraction methods mainly include visual interpretation,band calculation,spectrum supervision classification,decision tree classification,etc.There are problems such as requiring expert experience,low efficiency,low automation degree,weak generalization ability,etc.It is difficult to realize rapid monitoring of large-scale remote sensing water body information.Deep learning has become an effective way to extract remote sensing image information due to its "self-learning" characteristic of data features and more effective feature extraction and fitting for high-dimensional image data,which has strong applicability in multi-band remote sensing image information extraction.This paper innovatively introduces a dense connection structure into the Fully Convolutional Networks,which alleviates the problem of data low-dimensional feature loss,strengthens the network's ability to extract image features,improves the sensitivity to detailed water bodies in remote sensing images,and achieves better water body extraction effect.The Yangtze River(Anhui section)is selected as the research area,and the deep learning method is applied to the actual remote sensing water body extraction task to realize the automatic acquisition of water body spatio-temporal evolution and ecological shoreline change data,providing data support and technical support for the "Yangtze River Protection".The specific conclusions of this paper are as follows:(1)In this paper,the water body extraction method based on deep learning from remote sensing images can be successfully applied to the water body extraction task from general remote sensing images.According to the actual results,the water body extraction results of this method are basically close to the results of manual visual interpretation,which can meet the accuracy requirements of water body information extraction from general remote sensing images.The experimental results show that the Pixel Accuracy of the method in this paper is 96.3%,the Mean Intersection over Union is 91.1%,the Patch Miss Rate is about 0%,the Length and the Area accuracy are 95.8%and 98.5%,respectively,which are improved compared with the traditional NDWI method,spectral supervised classification method and decision tree method.In addition,in the experiment,the method presented in this paper shows extensive advantages that conventional remote sensing methods do not have,which means that the method breaks through the limitations of conventional remote sensing methods on remote sensing image sensors,spatial resolution,time,location and other factors to a certain extent,and in terms of method efficiency and automation degree,the method also obviously realizes automation,intelligence and integration of water body extraction from remote sensing images due to conventional remote sensing methods,greatly enhances the practicability of the method,and provides technical support for related water body research(2)The method of this paper is used to extract water from Landsat 8 OLI images in autumn and winter of 2013-2018 and Sentinel 2 images in rainy and dry seasons of 2018 in the Yangtze River(Anhui section)to obtain accurate water body and ecological shoreline data.According to statistics and analysis from the water area of the main stream of the Yangtze River and the length of the coastline in the north and south,it can be concluded that the change trend of the water area and the length of the coastline of the main stream of the Yangtze River(Anhui section)from 2013 to 2018 is basically the same from the interannual point of view,which shows that the area and the length of the coastline are basically stable except for the sudden increase in 2017 years due to the connection of the Wuhu section of the Yangtze River to the lakes along the river,,the maximum fluctuation range of the area is about 60km2,and the maximum fluctuation range of the length of the coastline in the north and south are 57km and 16km respectively.From the change of landscape index,from 2013 to 2018,the fragmentation and irregularity of water landscape in the 5km buffer zone of the Yangtze River(Anhui section)showed an overall upward trend.Judging from the changes in high and low water seasons,the area of the Yangtze River(Anhui section)in 2018 increased by about 92km2 compared with that in low water seasons.In high water seasons,the south and north bank lines increased by 20km and 25km respectively compared with that in low water seasons.The area and coastline were mainly caused by the changes in water levels along the tidal flats and river bank lines around the Yangtze River.In addition,the blocking of tributaries of the Yangtze River caused by the changes in water levels and the connection between the main stream and lakes along the river were also one of the reasons for the changes in the main stream of the Yangtze River.