Research On The Extraction And Classification Of The Main Features Information Of High-resolution Remote Sensing Image Based On Object-oriented

With the rapid development of remote sensing technology,more and more high spatial resolution remote sensing images are applied,which creates favorable conditions for obtaining a large number of ground information quickly and accurately.The traditional method of extracting remote sensing image information based on pixels mainly considers the spectral features of ground objects,ignoring the texture features and spatial features of target ground objects,resulting in the loss of image information.The appearance of object-oriented extraction method makes up for the shortage of pixel classification.This method takes image objects as basic units,divides image objects first and then classifies them.The segmentation process fully excavates the information of high-resolution images.The phenomenon of pixel misclassification and missing classification within the same ground object range caused by pixel classification method is effectively avoided,and the "salt and pepper phenomenon" is reduced.In this paper,the object-oriented technology of hierarchical multi-scale segmentation is studied.In the research process,Yalong Bay Resort in Sanya City,Hainan Province is taken as the experimental area,QuickBird image is taken as the data source,and object-oriented software eCognition is selected for image segmentation and classification.Finally,ideal results are achieved in classification accuracy.This paper mainly includes the following aspects:1.The quality of the fusion effect greatly affects the classification accuracy of the target features.In order to obtain high-quality remote sensing image,this paper compares PC Spectral Shaping,Brovey,Gram-Schmidt,NNDIffuse Pan Shaping four fusion algorithms based on envi software,and obtains NNDIffuse pan through comprehensive visual interpretation of subjective evaluation and calculation of objective evaluation in MATLAB Sharping is superior to other fusion methods in spectral fidelity and information richness.2.In order to get the best segmentation effect,this paper studies the parameters that affect multiscale segmentation:band weight,heterogeneity factor weight and segmentation scale.Esp2(estimation of scale parameter)is introduced to assist the acquisition of water area,vegetation,roads and buildings.The optimal segmentation scales are 250,190,102,60 respectively,which improves the efficiency of parameter acquisition.Combined with the characteristics and prior knowledge of the ground objects in the experimental area,the wave band weight,spectral factor weight and shape factor weight of the main ground objects are determined.According to the scale of segmentation,the hierarchical structure of main features is established.3.The main feature information is extracted and classified by optimizing the feature.In this paper,based on the in-depth study of the spectral characteristics,texture characteristics and spatial characteristics of the surface features,the main features of the experimental area:water area,vegetation,roads,buildings and different combinations of characteristics and threshold conditions are proposed.According to different objects,the corresponding rule set is established,and the main objects information is extracted by threshold classification and fuzzy classification,and thematic map is generated.The overall accuracy of classification is 91.3%,and kappa coefficient is 0.88.At the same time,it is compared with object-oriented nearest neighbor classification.The results show that this method is better than the standard nearest neighbor classification.