Research On Building Information Extraction Based On High Resolution Remote Sensing Image

The development of remote sensing technology is accelerating,and the resolution of acquired remote sensing images is getting higher and higher;the information contained in high-resolution remote sensing images is several times or even tens of times that of low-resolution remote sensing images.The feature information needs to improve the previous information extraction method.The efficient and accurate extraction of target features from high-score images is a hot issue in remote sensing information extraction at this stage.Buildings,as important feature information in remote sensing images,play a vital role in urban planning and digital city construction.Therefore,it is necessary to study a practical and effective information extraction method to classify and extract buildings.This paper takes object-oriented classification method to extract buildings in high-resolution remote sensing images as research content,and uses e Cognition as platform to extract object-oriented building information.The main research results are as follows:1.Introduced the current research status of high-resolution remote sensing image information extraction and the theoretical knowledge of related segmentation classification mainly involved in this article,and briefly introduced the principle and method of multi-scale segmentation.2.A method of building information extraction with the aid of shadow and edge detection is proposed.Multi-level segmentation and classification of remote sensing images is first performed.The shadows of buildings and other features are first extracted,and then the unclassified areas are constructed.Material extraction,which can improve accuracy.In the segmentation process,use the scale evaluation tool to select the optimal segmentation scale for segmentation.3.The improved Canny algorithm is used for edge detection of remote sensing images,and the improved algorithm optimizes the noise error in the traditional algorithm.And the results of edge detection are involved in segmentation,and the band weight is set to 3,which improves the classification accuracy.4.Use e Cognition software's multi-scale segmentation technology to filter the image segmentation parameters and homogeneity factors.The experimental results show that the optimal segmentation scale for shadows is 155,and the optimal segmentation scale for buildings is 205,when the segmentation scale is At 85,it is best for segmenting small features in an image.The segmentation effect is best when the shape factor and compactness of the homogeneity factor are both 0.5.5.From the accuracy evaluation results,we can know that the overall classification accuracy obtained by the classification method assisted by shadow and edge detection has reached 89.7%,the accuracy of the building has reached 98.43%,and the Kappa coefficient has reached 0.835.The classification result is much higher than the classification accuracy without using auxiliary conditions.In this paper,the advantages and disadvantages of threshold classification and nearest neighbor classification are also experimentally compared,and the nearest neighbor classification is more suitable for Level2 classification.