Research On Object-oriented High-resolution Image Multi-scale Segmentation And Land Use Classification

High spatial resolution image is an important direction for the development of remote sensing data.Accurate and efficient extraction of land use classification information by high-resolution remote sensing image is one of the hotspots in the field of remote sensing.The remote sensing-based land use classification information extraction has low precision and land use.With the complicated situation,the traditional method cannot meet the requirements well.The research uses the object-oriented high-resolution image classification method to study the selection of optimal parameters in hierarchical segmentation,and combines the two-level scale set segmentation with the Classification And Regression Tree,through the land use classification experiment and the pixel-like maximum Comparing the results of the method with the object-oriented closest method,the following results were obtained:Firstly,from the scale theory,the image segmentation method based on Fractal Net Evolution Approach algorithm and scale set model is integrated,and the Bi-level Scale Set model is constructed by using block segmentation and parallel computing optimization scale set model.The segmentation experiment is carried out on the object and the best selection is made.The segmentation parameters and the optimal segmentation method are constructed to construct the corresponding segmentation object layer,which effectively avoids the over-segmentation and under-segmentation problems that are common in hierarchical segmentation and improves the effect of hierarchical segmentation.Also,the image segmentation experiment is carried out by the maximum area method,the mean variance method and the Estimation Scale Parameter tool verification method,and the optimal segmentation parameters are selected,including: scale parameters,band weights,shape factors and compactness factors.And improve the segmentation edge extraction effect by constructing a segmentation auxiliary layer.Based on the optimal segmentation parameters and the segmentation results of the FNEA algorithm and the two-level scale set model,the segmentation effects of the two segmentation methods under the same optimal segmentation parameters are studied..Finally,The CART classifier is constructed,and the training samples and corresponding feature variables are selected to realize the automatic feature selection and threshold selection of the classification process.Trimming and optimizing the CART decision tree based on Salford Predictive Modeler,designing accurate and reasonable classification rules with spectral features,improving the accuracy of object-oriented classification experiments,realizing accurate and efficient land use classification and extraction,obtaining classification results and performing accuracy evaluation,The alignment analysis was performed using a pixel-based maximum likelihood method and an object-oriented closest method as a control experiment.The paper innovatively combines the Bi-level Scale Set segmentation with the CART,and has achieved good results in the land use classification research.