Research On Reed Beach Extraction Method In East Dongting Lake Based On GF1 Satellite Image

Wetland is known as the "kidney of the earth" and is closely related to the survival,reproduction and development of human beings.As an important component of the East Dongting Lake wetland,the reed beach has important functions such as adsorbing harmful substances,purifying water sources,and regulating climate.In the past,East Dongting Lake paper mills used reeds as raw materials to make paper,which can effectively maintain the normal growth of reeds.However,with the gradual withdrawal of the paper mills in the East Dongting Lake,a large number of reeds were abandoned,and reeds grew wantonly,blocking the rivers,and increasing the hidden dangers of floods.Therefore,it is very important to use appropriate methods to supervise them.With the development of remote sensing technology,the use of remote sensing images to extract and monitor wetland has gradually become the mainstream.In recent years,with the increase in the amount of remote sensing image data and the continuous improvement of the level of computer hardware,deep learning has gradually been applied to the extraction of wetland features from remote sensing images.This paper takes the images of East Dongting Lake taken by the GF1 satellite as the research object,and proposes two reed beach extraction models.The research contents are following:1)Considering that the problem of low extraction accuracy due to noise interference from water and vegetation in the extraction of reed beaches in the East Dongting Lake area,a reed beach extraction method based on multi-dimensional shallow features is proposed.First,shallow features such as water feature and vegetation feature are extracted to construct feature sets.Then,the random forest algorithm is used to optimize the features and the multi-dimensional shallow layer features are obtained.Next,a two-stream network model is built.The multi-dimensional shallow features are used in the coding layer to guide the module to make full use of the information in the multi-dimensional shallow features.Finally,through decoding layer,the features after fusion are up-sampled to obtain the extraction map of reed beach.Experiments show that the comprehensive extraction effect of this method is better than other comparison methods.2)In view of the problem that the characteristics of reed can only be reflected from one side by the single shallow layer,which leads to the incomplete extraction of reed beach,an extraction method of reed beach based on multi-feature recalibration is proposed.This method combines multiple shallow features in the network coding layer at feature level.Channel optimization and position weight redistribution are carried out for the fused features.Then the recalibrated features are used to guide the original data to extract the reed beach.Experiments show that the multi-feature-based recalibration method proposed in this paper can significantly improve the accuracy of reed extraction.3)Based on the multi-feature recalibration reed beach extraction model proposed in this paper,using a client/server design model,under the Py Qt5 application framework,based on Python 3.6 language this paper design and implementation of remote sensing image reed beach extraction software.Finally,software-related functions were tested on the multi-spectral remote sensing data in the East Dongting Lake region of Hunan Province.