Research On Wetlands Secondary Classification Based On Remote Sensing Images

The use of remote sensing technology to monitor wetland areas has become an increasingly important means to address some scientific issues in wetland research,such as wetland type information,wetland landscape information,and wetland change characteristics.This article mainly combines spectrum,texture and other characteristics,mainly using the classification method of support vector machine to complete the classification of medium and high-resolution remote sensing images,wetland information extraction.First of all,in the image information classification method,the method and flow of the support vector machine(SVM)image classification method in the wetland surface coverage information was studied,so that the surface coverage information data can be obtained more quickly and efficiently.Secondly,support vector machines have better performance in the training and classification of small samples.The reason is that the support vector machine's dynamic adjustment kernel function can adapt to different classification objects and conditions and has certain algorithm advantages.In the first-class classification process,using support vector machine classification methods,multi-spectral,texture and other features are added to classify,and the accuracy of the classification results is evaluated by using the precision evaluation reference method of ENVI in remote sensing image processing software.The results show that based on the support vector machine method experiment,the accuracy of the error matrix is used to obtain the classification accuracy of wetland remote sensing classification in the study area.The overall accuracy of the classification is 95.89%,and the Kappa coefficient is 0.93.Afterwards,we continue to use support vector machine classification(SVM)to perform secondary classification of wetland information.The overall classification accuracy obtained is 93.5%,and the Kappa coefficient is 0.89.However,the support vector machine classification method is not suitable for multi-sample remote sensing images.Further experiments were conducted on this issue.The SVM method and object-oriented method were used to extract information about cultivated land,green land,and wetland in the experimental area data.The results of information extraction by different classification algorithms and their classification effects were measured and compared.When the samples were gradually increased and classified,the accuracy of the support vector machine was gradually reduced and the magnitude of reduction was greater,while the object-oriented method was also available.It is reduced.but it can also maintain accuracy above 90%.Through experiments,when the range of small samples is determined,the overall accuracy of the SVM method in extracting wetland type information is higher than that of the object-oriented method,and the accuracy can be maintained at more than 90%;then,after the sample is gradually increased,the object-oriented method is used.The classification results obtained are obviously better than other classification methods,and can achieve better classification accuracy.As an important method in remote sensing image classification research,support vector machines(SVMs)successfully extracted the wetland information needed in the images and analyzed and evaluated the experimental results.This proves that the proposed support vector machine is used for wetland ?.The effectiveness and feasibility of the classification method.