Object-Oriented Classification Based On FNEA And Its Applicationin High-Speed Railway Line Extraction

In the process of object-oriented classification for remote sensing images,avoid the influence of "Same images different object " and "same object different images " phenomena on the image classification effect.It canalso create better conditions for high-resolution remote sensing imagetoextract features.This article uses the specific application of remote sensing imagery feature extraction as a background,reviews and summarizes the research progress and status quoof the domestic and foreign research.Through the introduction of the existing classifiers,take the image segmentation in the object-oriented processing of remote sensing images as a key content to furtherstudy.Object-oriented image processing usually encounters the problems of segmentation methods and segmentation parameters.The classification results will be better by taking experiments repeatedly,but the problems such as inefficiency and low reproducibility still exist.Therefore,theuseage of segmentation parameter optimization tools or algorithmpredicts the segmentation parameters of image,which is an effective method for object-oriented classification processing.In view of the above,this paper studies the effect of different segmentation methods and different segmentation parameters on the image in the object-oriented image segmentation process,and classifies the segmented images by Support Vector Machine(SVM),Classification And Regression Tree(CART)and K Nearest Neighbor(KNN)three object-oriented classifiers.Finally,the object-oriented remote sensing image classification method is applied in the example of high-speed railway line extraction.The main work and results are as follows:(1)Introducing six different multi-scale segmentation methods: Mean Shif segmentation,Watersheds segmentation,checkerboard segmentation,quadtree segmentation,Fractal Net Evolution Approach(FNEA)segmentation,and spectral difference segmentation.the comparison shows that FNEA multi-scale segmentation method has good applicability in different remote sensing image segmentation process.(2)Introducing the mean and standard deviation of the image spectral features,the shape index of the geometric features of the image.Also evaluating the image segmentation by the segmentation time,the number of segmented objects,and the maximum area of the segmented object,describing the trend of changes in these indicators with scale conversion under different segmentation parameters,summarizing the segmentation result.We find that the image segmentation effect is better when the scale parameter around 30-60,which can reflect the difference between different features;(3)Segmentation parameters prediction of IKONOS remote sensing image through ESP tools,setting different comparison parameters and prediction parameters for object-oriented classification and comparison,experimental results show that the ESP-based segmentation parameter optimization method is feasible in the object-oriented classification process of remote sensing images,It can reflect the image segmentation parameters more directly,alsoimprove the efficiency of object-oriented image classification processing to some extent.In addition,among the three classifiers in this paper,object-oriented image classification based on KNN classifier has better generalization,the classification effect is much higher than that of CART and SVM classifier under the same parameters.Meanwhile,the classification effect of SVM classifier is relatively low;(4)Based on the previous study and the summary of the experimental results,the specific application of the object-oriented optimal scale selection tool and KNN classifier for the high-speed railway line extraction,qualitative and quantitative evaluation of the experiment can be concluded that the object-oriented image classification method has certain advantages in the process of extracting specific features of remote sensing images.