Research On Building Extraction Technology Based On Multi-source High-resolution Remote Sensing Data

In recent years,space remote sensing technology has developed rapidly with the rising demand for spatial information of remote sensing.Mainly for the improvement of spatial resolution of remote sensing data and the enrichment of ways that can gain remote sensing data.The improvement resolution of the remote sensing data can enhance the ability of terrain information acquisition also bring interference because of the more complex terrain category,and it may bring the issue that "different objects with same spectrum" due to the decrease of spectral band because of the limit of the sensor restrictions,these problems has brought new challenges for information extraction of high resolution remote sensing image.With the development of multi-source remote sensing technology,remote sensing data like LiDAR are easier to be acquired,it is more advantageous to the extraction of target information by combining the merits of different remote sensing data.As an important apartment in remote sensing image,building is closely related to human activities.How to quickly and easily extracted building from high resolution remote sensing image has been the primary issue in city planning,military detection and other application about building.Firstly,this paper described the characteristics of the high resolution remote sensing image and digital surface model(DSM)data and introduced the preprocess methods of related data.And researched two different buildings extraction methods according to the presence of supervision information.According to the merits and demerits of the two building extraction algorithms,we studied how to effectively combine different remote sensing data,the use of different data sources for the features describe the characteristics of different buildings,combined with analysis and extract the effective information,the structure of the extracted results are obtained.For the unsupervised situation,this paper proposed a building extraction method based on texture overlap model(ORT).Firstly,using ORT method to get segmentation and extraction of building in a similar areas.Then,using the OTSU method to get the threshold segmentation of DSM grayscale image,getting the elevation regions after segmentation.Lastly,we can get more complete and clear building area with comprehensive two parts.The simulation proves that the proposed method can effectively extract the construction area.For supervised situation,this paper put forward a method to extract building based on support vector machine(SVM).We firstly get the fusion of the spectral data and DSM data,then get the low-level feature extraction from the fusion data,including spatial neighborhood features and texture features,etc.Then obtain the high-level semantic features of buildings through SVM,get the fusion of high-level semantic features and the original low-level features,using SVM to extract buildings again.Lastly,using morphological filter to optimize the extraction results.The simulations show that the building extraction method that combined with the DSM data and high-level semantic features has a improvement on accuracy and completeness.Lastly,simulation on three-dimensional reconstruction of building extraction results show that the proposed building extraction method has important significance of application on terrain reconstruction.