Study On Extraction Of Wetland Type Based On GF-1

Currently wetland resource is considered to be important the same as forest and marine ecosystems and thus its dynamics and sustainable protection and utilization is of a major concern.Using remote sensing technology to monitor wetlands has become a more and more powerful tool,which can solve some scientific problems we are facing in wetland research,such as wetland classification,wetland landscape mapping,wetland change detection and so on.Without doubt,the wetland study will promote the development of wetland ecological resource protection and therefore provide useful information for government's scientific and correct decision making in wetlant protection and reservation development.In this study,East Dongting Lake was selected as the research area,and the GF-1 remote sensing image was used as the data source.Firstly,the GF-1 data was preprocessed,and then the best band combination and the best fusion method were analyzed.So that we found the best combination of band and the best fusion method for GF-1 image wetland classification.Then,the genetic algorithm and Fisher discriminant method is introduced into the wetland extraction in the East Dongting Lake wetland type extraction,a optimization algorithm of support vector machine to achieve high precision extraction of wetland types,on the other hand,using the Fisher discriminant method of wetland types were simple,efficient and automatic extraction;at the end of the four algorithms to analyze and evaluate the results the aim is to analyze the optimal algorithm,applicable to GF-1 remote sensing image extraction of wetland types,improve wetland type extraction system,provide the basis for future wetland remote sensing research.The main findings are as follows:(1)The best band combinationIn this paper,we considered both spectral characteristics and the amount of information,obtaining the best band combination of GF-1 image(RGB=432),which is determined by quantitative evaluation indexes including standard deviation,information entropy and best index.(2)The best fusion methodThe fusion results were evaluated by quantitative analysis,including the spectral inheritance and spatial integration.The mean,correlation coefficient,entropy,standard deviation and gradient were used to compare three kinds of fusion methods,namely PCA(Principal Component analysis),GS(Gram-Schmidt Pan Sharpening)and SFIM(Smoothing Filter-based intensity Modulation).SFIM fusion method is superior to other methods in spectral inheritance and spatial integration.It can improve the spatial resolution and retain the spectral information well,which is beneficial to the information extraction.So SFIM is a good fusion method for GF-1 image.(3)Support Vector Machine Optimized by Genetic AlgorithmThe overall accuracy of the support vector machine and the support vector machine optimized by genetic algorithm are 83.79%,88.14%,respectively.The kappa coefficient is 0.7985,0.8527,respectively.They have a difference of 4.5 percentage points,and the kappa coefficient is different from 0.0542.The extraction time is basically the same.It is shown that the support vector machine optimized by genetic algorithm is effective in the extraction of wetland type,and the extraction precision is obviously improved.(4)Fisher discriminant automatic extractionThe convergence of Fisher discriminant is much improved,the number of data iterations is significantly reduced,the extraction speed is significantly improved,and the time of extraction is only 49 seconds on the wetland type extraction of GF-1 image.The Fisher discriminant method has obvious advantages in large volume image processing,which can meet the requirement of overall precision and shorten the extraction time of wetland.(5)Comparison of wetland type extraction algorithmsThe results show that the support vector machine and the object-oriented decision tree with the best accuracy are 88.14%,followed by the overall accuracy of the Fisher discriminant method by 85.17%and the support vector machine with the lowest overall accuracy of 83.79%.The Kappa coefficient of the support vector machine is 0.8572,followed by the object-oriented decision tree and the Fisher discriminant method,which has the Kappa coefficients of 0.8217 and 0.8129 respectively,and the Kappa coefficient of the support vector machine is the lowest,which is 0.7985.This shows that the overall accuracy of the support vector machine(SVM)is better than Fisher discriminant and support vector machine,and the Kappa coefficient is much better than the other three extraction methods.Fisher retrieval method has the shortest extraction time(49 seconds),followed by object-oriented decision tree(165 seconds),genetic algorithm optimization support vector machine and support vector machine have the longest extraction time(253 seconds,249 seconds,respectively),which shows Fisher discriminant in wetland type Extraction time is superior to the other three classification methods.