Research On Building Extraction Based On Object-oriented High-resolution Remote Sensing Image

Buildings as the main geographical elements of a city,whether it is from the national development level or urban development planning,it is very important to accurately grasp the spatial location and shape information.The high spatial resolution peculiar to high-resolution remote sensing images makes the detailed information of features more fully expressed,thus causing the bottleneck of traditional pixel-based features recognition and extraction methods.Therefore,this paper takes buildings as a research goal,and uses object-oriented classification technology as a research method,comprehensively applies the spectrum,texture,and geometric characteristics of buildings on high-resolution remote sensing images,and builds on object-oriented high-resolution remote sensing images.Research on bio-extraction.The main research contents and results of the article are as follows:(1)In order to improve the over-segmentation phenomenon caused by the classic watershed transformation method,this article is based on the watershed algorithm in the area segmentation method,and mainly improves it from two aspects:First,it combines the Canny edge detection algorithm to obtain a gradient image with prominent edge information;The second is to find the labeling seeds for area labeling in combination with distance transformation.(2)Use the result map of the improved watershed algorithm to assist multi-scale segmentation to achieve the optimization of multi-scale segmentation technology.Through experimental analysis,the optimized multi-scale segmentation method has the characteristics of Canny edge detection and improved watershed algorithm.This method can clearly and accurately express the edge information of the building and effectively reduce the over-segmentation phenomenon in multi-scale segmentation.(3)Aiming at the problem of "dimensional disaster" in the process of object-oriented remote sensing image classification,this paper proposes a feature selection algorithm combining Relief F algorithm and Binary particle swarm optimization(BPSO),using data The analysis platform realizes the selection of the optimal feature subset,and finally selects 10 optimal classification feature subsets for classification extraction from the 63-dimensional original feature set.(4)Based on the optimized multi-scale segmentation technology proposed in this paper and the Relief F-BPSO feature selection algorithm,the nearest neighbor algorithm provided by the image analysis platform is used to realize the object-oriented high-resolution remote sensing image building classification extraction.And the application of the image analysis platform's own editing function on the classification extraction results was used to fine-tune and precision verification of the feature category.Through the experimental comparison with the nearest neighbor method based on multi-scale segmentation and empirical feature selection,the extraction scheme of this article was obtained The user accuracy,producer accuracy,overall accuracy and Kappa coefficient have been increased by 2.7%,15.4%,9.8%,and 12.4%,respectively,indicating the feasibility of this scheme.