Object-oriented Classification Method Based On Rule Verification Points And Its Application In Building Extraction

In the development of remote sensing image classification technology,the use of image surface feature information,such as spatial features,spectral features,texture features,etc.has become the main method of image classification processing.Initially,people proposed to use spectral features as the main method of pixel-based supervised classification and unsupervised classification.However,in the classification process,they often bring about the phenomenon of "same spectrum" and "same spectrum".In order to make full use of the image surface feature information and effectively improve the classification accuracy,an object-based object-oriented classification method is proposed,which reduces the phenomenon of "same objects with different spectra" and "same spectrum with foreign objects" in classification.However,the post-classification processing commonly used in object-oriented classification methods ignores the errors caused by random verification points in the past,thereby reducing the post-classification accuracy.To this end,this paper proposes an object-oriented classification method based on rule verification points.At the same time,the object-oriented classification method based on rule verification points and the median absolute deviation(MAD)algorithm are combined to optimize the training samples in object-oriented classification.The experimental comparison through two sets of data shows that: rule-based verification The point MAD-KNN classification method has a significantly improved accuracy than the original KNN classification method.In this paper,the main research contents are as follows:1)This article mainly studies the development status and method applicability of image segmentation methods.Introduce multi-resolution segmentation,quadtree segmentation,Mean-shift segmentation,Watersheds segmentation and other segmentation methods.Through the comparison of renderings and standard deviation,pixel number,mean,maximum area and other data,it is concluded that the multi-resolution segmentation method has a strong Applicability.2)This article mainly applies ESP tools to object-oriented classification,which improves the classification efficiency.The ESP scale selection tool is used to predict the optimal scale parameters of the GF-2 image.The reasonable setting of the data around the prediction parameters is combined with the image segmentation method to compare the feasibility of the ESP tool.The results obtained are compared with the results of three representative object-oriented classification methods.The experimental results show that the ESP tool effectively improves the image classification efficiency.Among them,the KNN-based object-oriented classification method has high applicability.3)In order to solve the problem of the difference between the result evaluation of the object-based object-oriented classification method and the actual classification effect,the article uses rule verification points to optimize.A method for evaluating accuracy results based on rule verification points is proposed.The comparative analysis of a large number of experimental results shows that the method proposed in this paper has a greater improvement than the traditional random verification point accuracy evaluation results,which effectively improves the accuracy evaluation results of image classification post-processing.4)For the problem of sample outliers in the process of object-oriented classification,the paper uses MAD algorithm to optimize.This paper uses the median absolute deviation algorithm to improve the object-oriented nearest neighbor(KNN)model based on rule verification points.Using the differences in pixel values between different features,we can effectively eliminate abnormal samples in the classification process.Experimental results show that the algorithm in this paper effectively improves the accuracy evaluation results while ensuring the accuracy of the samples.Aiming at the problem of instability based on rule verification point MAD-KNN method under different building density.In this paper,the MAD-KNN method and the object-oriented morphological building index(MBI)are used to compare the building extraction of remote sensing images under the same conditions.The experimental results show that the object-oriented MBI method is more prominent in the extraction of high-density buildings.Highland suitability.