Font Size: a A A

Study On Decision Tree Method For Land Use Information Extraction Based On Remote Sensing Images

Posted on:2014-10-05Degree:MasterType:Thesis
Country:ChinaCandidate:J Q ChenFull Text:PDF
GTID:2250330428959611Subject:Soil science
Abstract/Summary:PDF Full Text Request
There is severe contraction between human being and land use in Chi-Shui river basin, and the unreasonable land use methods seriously impact on the sustainable use of the basin land resources and protection of the ecological environment. Keep abreast of the current land use condition of each sub-basin of Chi-Shui river basin is of great significance for carrying out of reasonable land use analysis, water and soil erosion control, and ecological and environmental protection work. Exploring a land use information extraction method within the study area to make the accuracy and efficiency unified can provide important support for river basin integrated management.Chi-Shui river basin was chosen as study area for this research, and ten ALOS multi-spectral remote sensing images that covered the region were used as data source. Based on the coverage of each image, the whole basin was divided into ten sub-basins and one of them was selected as test area. The image spectral characteristics, Normalized Difference Vegetation Index (NDVI), Normal Differential Water Index (NDWI), Digital Elevation Model (DEM), and eigenvalues’numerical difference after image band computing of the eight typical land use types were fully analyzed. Based on these analyses, the thresholds for differentiating land use types were set, and a binary decision tree model which base on the rules of threshold was explored and established for land use classification. The method and idea were then expanded to the remaining nine sub-basins, and the factors affecting the classification accuracy of different image were statistically analyzed.The main findings are as follows:(1) A rule-based decision tree model for the No.l sub-basin (test area) that is geographically representative for the whole basin was built. Based on this model, the overall accuracy of eight land use type classification results is89.05%, the Kappa coefficient is0.8741. Compared to the maximum likelihood method and support vector machine method, the overall accuracy of this method is higher by12.39%and10.78%, respectively; The Kappa coefficient of this method is higher than the maximum likelihood method and support vector machine rules by0.1412and0.1238, respectively. Overall, the model more or less reduced misclassification errors of classification results of woodland, grassland and shrub, river, reservoir and pond, terrace land, paddy field, road and construction land. Among the seven land use types, the decrease rates of (1) rivers and (2) reservoirs and ponds are most obvious, of which maximum decrease were50.25%and46.71%. All in all, these results show that this method has good applicability and operability in the No.1sub-basin.(2) The successfully tested methods and ideas were then expanded into basin-wide ten sub-basins and the overall accuracy and Kappa coefficient of rule-based decision tree classification results in eight sub-basins are significantly better than the ones of maximum likelihood method. The highest overall accuracy of classification results is90.59%and the Kappa coefficient is0.8811, which are higher than the ones of the maximum likelihood method by7.4%and0.0887. These results show that the rule-based decision tree method has a certain degree of universality in Chi-Shui river basin and it can effectively improve the land use classification accuracy.(3) According to the156GPS field measurements points within Chi-Shui river basin of this study, a database was established after differential correction, and the accuracy of basin-wide land use map classification results was validated. The validated results show that133in156points land use classification are correct and the validated accuracy of field measurements points is85.26%.(4) The obtaining time of remote sensing data influences the decision tree model buiding. Statistically, based on the multiple linear regression model analysis, it shows that the Kappa coefficient of the ten sub-basins is inversely proportional to the total area of the classification result, and is proportional to the area of paddy fields, which represents that the higher the total area of an image of the sub-basin, the lower classification accuracy it has and the higher the paddy field area of an image of the sub-basin, the higher the classification accuracy it has.
Keywords/Search Tags:ALOS multispectral image, Decision tree, Land use, Chi-shui riverbasin
PDF Full Text Request
Related items