Remote Sensing Water Extraction And Research Based On Improved U-Net Model

Water resources are the key to maintaining the prosperity of human economy,the stability of production and development,and the sustainable development of ecological system.With the rapid development of remote sensing technology,real-time updated remote sensing images provide an effective data source for surface water body identification.Early methods of surface water extraction were based on spectral information and classifier methods to build theoretical models,which were difficult to apply in areas with complex and variable water types.Depending on the powerful computing power of the computer,in-depth learning can automatically recognize objects from remote sensing images by fitting unknown data to the learning of prior samples,which is favored by researchers.However,the existing in-depth learning of water body recognition network model is inefficient in water body extraction tasks of large-scale and high-resolution remote sensing images.Therefore,this paper presents an efficient water body extraction network model based on improved U-Net remote sensing image.Based on the results of water body extraction,the spatial and temporal distribution of water body changes in Liaoning Province was monitored,and the influencing factors of water body changes were revealed,which provided important reference for agricultural production,water conservancy planning,natural disasters and other decision-making.The main work and contributions of this paper are as follows:(1)Make large water datasets.Existing remote sensing imagery semantically segmentes water bodies with less open datasets,low resolution and single background.This paper uses GF-2 data of 1 m spatial resolution after fusion of multispectral and panchromatic band images and entinel-2 data of 10 m spatial resolution to make a dataset.After refined labeling,the data is enhanced to increase the sample size of the dataset and improve the generalization ability of the dataset.It can provide data support for water distribution and dynamic change law.(2)A water extraction network model based on improved U-Net remote sensing image is presented.This model simplifies the network by reducing the number of layers of VGG16 backbone network,and improves the efficiency of U-Net based on the accuracy of feature extraction.Experiments on two high-resolution remote sensing image data,GF and Solidnel,show that the improved model has excellent segmentation performance and speed,and can meet the needs of water extraction in a wide area.Taking the result of water body of GF image as an example,the accuracy of IOU and above evaluation index is 90.311%,94.909%.The training duration was reduced by 25.8% and the predicted duration by 21.6%.(3)A water body classification method based on shape index is constructed,and the water body changes of long-time remote sensing images are analyzed according to the classification results.This paper classifies water bodies based on shape index,and achieves fast acquisition of geometric features and categories of water bodies.The Autumn Sentinel-2 remote sensing images from 2017 to 2021 in Liaoning Province were selected as data sources to analyze the area changes of different types of water bodies.At the same time,combining with the data of air temperature,precipitation,potential evapotranspiration and human activities in various cities of Liaoning Province over the past five years,the trend of change is fitted and analyzed,and the following conclusions are drawn:(1)From 2017 to 2021,the water area in Liaoning Province showed a decreasing trend as a whole.The total area of water decreased by 2.16% during the study period.The area of all three types of water bodies decreased,among which the proportion of small and fine water bodies was larger,which was 62.45% of the total area reduction.(2)The change of water area in Liaoning Province from 2017 to 2021 is the result of the combined action of natural climate and human activities.Evapotranspiration from climate factors is the main driver of water area change in the region,and human activity is the secondary driver.(3)Due to the double effects of global warming and urbanization,the evolution law of surface water bodies in Liaoning Province has changed significantly,which will affect the protection and repair of regional ecosystems and hinder the high-quality development of national economy.Therefore,it is necessary to formulate a plan for ecological protection and recovery in Liaoning Province to prevent further degradation of the ecological environment in some parts of Liaoning Province,and to balance the relationship between human development and ecological system protection and recovery.