High-resolution Remote Sensing Image Building Extraction Research

In recent years,due to the rapid development of remote sensing technology,it has promoted the increase of the spatial resolution of remote sensing images.Detection and extraction of artificial features from high-resolution remote sensing images has become an internationally forefront topic in the field of remote sensing applications,image processing and computer vision,which has significant theoretical and practical value.Geographic information updating system,urban management and planning,ecological environmental monitoring,map navigation,disaster relief and agriculture field all rely on high-resolution remote sensing image building extraction.However,there are still many difficulties in practical application of high-resolution remote sensing image building extraction.There is still a lot of research needs to be done in order to make it perfect and mature.In the thesis,aiming at some problems in the existing building extraction methods,we use high-resolution remote sensing images as data source,establish a basic building model from the geometric characteristics of the building,the corresponding building extraction method is given,and the building outline is normalized and extracted.The content of the thesis including:1.The most disturbing factors of the building extraction are the vegetation with the similar height adjacent to the building and the road with similar texture and color to the building.Based on this,the thesis takes two steps before the building extraction.First,a method of extracting vegetation is proposed,and the vegetation likelihood model is obtained by offline learning as prior knowledge.Second,introduce FLD(Fisher Linear Discriminant)and shape feature to recognize road information,and use the knowledge of morphological to optimize the road network to achieve the purpose of road information removal.These two steps provide favorable conditions for the subsequent building extraction and reduce the interference of background information to the building.2.The most commonly used neural network algorithm in machine learning and artificial intelligence is introduced into the field of building extraction.In this thesis,an algorithm of building recognition and extraction based on LVQ neural network is proposed.Through experimental simulation,it is proved that the algorithm has good accuracy and adaptability.3.To study the calculation of the building's main direction,which characterizes the orientation of the building.Therefore,the thesis aims at how to calculate the main direction of the building accurately and quickly,proposes an improved straight line fitting method and a new main direction estimation algorithm based on ISOCLUS clustering.4.The right-angled polygon building model based on rectangular structure is introduced.Based on this building model,the algorithm of building outline extraction based on the polygon fitting is proposed.The experimental results show that the algorithm can effectively realize the regularized processing of high-resolution remote sensing image building outline extraction with all kinds of complex shapes.