Classified Study To High-Resolution Remote Sensing Image Of Wetland Coverage

High-resolution remote sensing images have replaced traditional low-resolution satellite remote sensing images because of their rich spatial information and texture information.For high-resolution image classification methods are also increasingly updated,people continue to explore the rich information in high-resolution images,making high-resolution satellite imagery more widely used in land resources,ENVIronmental monitoring,urban planning,precision agriculture and other fields.This paper firstly performs image preprocessing and image fusion on the original high-resolution remote sensing image of Shifosi wetland.Using object-oriented classification,supervised classification,unsupervised classification,and decision tree classification based on expert knowledge,four classification methods based on Word View-2(2011)and GF-1(2018)high-resolution satellite imagery of two years were used to analyze the classification of small-area wetland land images according to classification accuracy.The best classification was screened.Then,using the best classification method,that is,the object-oriented method,the area after the classification is statistically evaluated,and the dynamic change of the land in Shifosi wetland is evaluated.The main research conclusions are as follows:(1)HSV transformation,Gram-Schmidt transformation,Brovey transformation and NND transformation experiment in ENVI,analysis of the fusion image,calculation of fusion evaluation index based on IDL programming,and finding that Gram-Schmidt transformation is in space Both information and texture information have high retention;(2)for the effectiveness of high-resolution image exploration classification methods,explore the classification results of traditional classification methods,supervise classification based on maximum likelihood method,using K-Means The method implements unsupervised classification,and the decision tree classification is based on e Cognition image segmentation for decision tree classification.In the object-oriented classification,the expressions of images under different segmentation scales and segmentation parameters are analyzed,and the optimal segmentation scales are selected for classification of buildings,water bodies,woodlands,artificial pastures,roads,tidal flats,reeds,lotuses and pupa.The mean variance and maximum area of ??the scale,the optimal segmentation scales of buildings,woodlands and roads in Worldview-2 and GF-1 are 120 and 80 respectively,and the optimal segmentation scales of reed and pupa are 240 and 140,respectively,lotus and grass.The optimal segmentation scales are 400 and 180,respectively,and the optimal segmentation scales for bare land and water are 580 and 200,respectively.(3)Calculating the accuracy of different image classification results of Word View-2 and GF-1,the classification accuracy of the two images is 88.13% and 90.8%,respectively.Unsupervised classification,supervised classification accuracy is less than 80%,decision tree classification accuracy is high,Word View-2 and GF-1 are 82.8% and 84.12% respectively;statistical analysis of object-oriented area,analysis of wetland vegetation based on single-land dynamic model The dynamic change of the type showsthat the growth rate of Pucao is the fastest,and the annual change rate of 2011-2018 reaches 11.58%.The second is the lotus flower,with an annual growth rate of 5.2% and a low growth rate of reeds(1.32%).The area of artificial grassland and forest land changed little,and the area of wetland vegetation in Shifo Temple continued to expand,creating a good regional ironment.