Research On Semi-supervised Classification Of Hyperspectral Image Based On Graph

Geographic information system is a comprehensive information management system,appearing with the development of computer technology.One can achieve collection,management,processing,analysis and modeling for complex geospatial data.As the best spatial data management tool,it reveals the characteristics of geographical data on space,attributes,time,etc.Spatial data mining is an emerging technology dedicated to data analysis,understanding,mining implicit relationships within data,and it has unique advantages in dealing with a large number of spatial geographic data.With the unprecedented increase in the number of geographic data,spatial mining technology can effectively solve the problem caused by the large amount of data.The combination of geographic information system and spatial data mining,it not only enables geographic information system breaking through the limitations of its own function,but also broadly extends the data mining technology in the processing of data types.Hyperspectral remote sensing has become the frontier in the field of remote sensing,Hyperspectral images classification is a high theoretical basis and has profound research value.This paper first introduces the relationship between the geographic information system and the hyperspectral remote sensing image classification technology,development situations and basic functions,Then summary on the research of semi-supervised classification,and next introduces several classic algorithms of semi-supervised classification is presented.Lastly,the several classic algorithms of semi-supervised classification based on graphs as well as their advantages and disadvantages.On this basis,the following research work is carried out on the semi-supervised algorithm based on graph:This paper presents a semi-supervised classification method for hyperspectral images based on spatial features and texture information.Firstly,the hyperspectral image is processed for dimension reduction by PCA.Then,the spectral characteristics,spatial characteristics and texture characteristics of the image are fused.Finally,the semi-supervised local and global uniform algorithm(LGC)is used to classify it.The experimental data are based on the Indian Pines dataset and the PaviaU dataset.The experimental results show that the classification results by combing spectral characteristics,spatial features and texture features of hyperspectral image are superior to those of single spectral characteristics.The use of texturefeatures leads to have a raise of the accuracy of 3%-4%,With the increase of the size,the classification accuracy has been significantly improved,All those prove the effectiveness of this method.