An Automatic Detection Method Of New Buildings On High Resolution Remote Sensing Images

With the acceleration of urbanization in China,land cover has changed rapidly.Its main effect is that the city is gradually expanding into the countryside and suburbs,so the urban building renewal and the detection of illegal buildings have become an important task for the land resource management department.Due to frequent changes in land cover,there is an urgent requirement for high-efficiency,large-scale automatic acquisition of dynamic building changes.The remote sensing technology and digital image processing methods have made great progress in recent years,and it has provided a technical guarantee for the automatic discovery of new buildings and the realization of large-scale dynamic detection.At present,the feature of using remote sensing images to detect building changes is relatively simple,which does not express the characteristics of buildings well,and the integrity and accuracy of identifying buildings are low;secondly,identifying building changes still requires the selection of training samples and manual intervention,contributing to low automation.In response to these problems,this study proposes an object-based new building automatic detection method for high-resolution remote sensing images.The paper mainly explores the following two aspects:(1)In this study,an automatic detection method for new buildings based on multiple features was proposed,using MBI(Morphological Building Index),Pan Tex and Harris features to extract new building information.After the feature condition is defined at the feature level,we use weighted difference and window semi-overlap to calculate the change hot spots(groups of buildings)of single features.(2)The object-based method is used to fuse the feature results.After the change hot spots is binarized and connected,the transformation information is converted from pixels to the object level.Then,the feature fusion is realized by object voting method,and the results are optimized by spectral constraints,shape constraints and morphological processing methods to complete the extraction of new buildings.In this study,the algorithm was tested in 23 counties in China,and the omission rate was below 32%,the average precision rate was 72.93%,and the average accuracy F1 ranged from 0.70 to 0.84,indicating that the algorithm has a relatively stable accuracy performance,verified the effectiveness of the algorithm.In order to explore the accuracy differences caused by different feature combinations,three experiments of MBI,Pan Tex and their object voting were set up.The results showed that the multifeature combination method had higher accuracy.The main reason is that the object voting of MBI and Pan Tex features can improve the recall rate of detection results,and Harris feature can further improve the precision rate,and finally achieve the improvement of average accuracy.In order to test the influence of feature fusion method on accuracy,direct feature superposition and change vector analysis were added as comparative experiments,and the results showed that multi-feature object voting had more advantages.The multi-feature object voting is based on MBI features,with Pan Tex and Harris features as auxiliary information,and the feature fusion strategy with different weights can avoid the low precision caused by the direct accumulation of multi-feature error information.This study proposes a new building automatic detection method for highresolution remote sensing images,which can realize automatic detection of new buildings without the need for training samples,greatly saves manpower,and provides strong technical guarantee for realizing rapid dynamic detection of land cover.