Automatic Building Extraction Of Remote Sensing Images Based On Fused Multi-Scale And Multi-Feature

Urban digital construction is developing rapidly.Buildings as one of the most important geographical data categories in urban geographic data,are the thematic elements that need to be highlighted in urban large scale mapping.The recognition and extraction of buildings are of great significance to feature extraction,feature matching,image interpretation and cartography.The information of buildings in remote sensing images can provide important decision-making support for land management,land use,urban planning,earthquake disaster monitoring and so on.It also has important applications in many fields,such as the construction of smart city,investigation of illegal buildings,and military reconnaissance.Therefore,automatic extraction of building targets from remote sensing images is one of the important research topics in remote sensing image interpretation.However,there are great differences in the size,shape,color and orientation of buildings in remote sensing images,which pose a great challenge to building detection.For this reason,this paper proposes an algorithm for automatically extracting buildings from remote sensing images based on multi-scale and multi feature fusion.In remote sensing images,buildings are displayed in the form of roofs.However,the color,texture and shape of the roof of the buildings are different,and the single feature is difficult to express all the buildings.Therefore,this paper presents a variety of effective characteristics to describe buildings,including edge density and edge distribution,brightness contrast,color contrast,which describe the bottom features of the edge,brightness and color of the building.At the same time,the characteristics of the main direction orthogonality,the integrity and the symmetry of the target are added,combined a variety of features to describe buildings together.Then use the Adaboost algorithm to train a series of samples,and the weights of various features in the multi feature model are obtained through iterative learning,thus the multi feature linear model is obtained to calculate the visual significance of the buildings in the sliding window.In view of the different size of buildings in remote sensing images,such as a factory of thousands of square meters,and a shed of few square meters,so it is difficult to show strong significance in a single scale window.In this paper,by constructing the Gauss Pyramid model,the fixed scale sliding window corresponds to the different actual ground area in the different layers,and the building saliency of the sliding windows of different levels is calculated by the multi-feature model to realize the synchronous detection of different scale buildings.At last,the visual significance of superpixel is calculated with the unit of superpixel,and the image is divided into buildings and background by Otsu threshold segmentation algorithm,and the building is extracted.In this paper,several groups of high resolution remote sensing images are used to experiment on the proposed algorithm,and the Markov random field model algorithm and the FCN algorithm are used as the contrast algorithms.The results of the experiment are compared qualitatively and quantitatively.The results show that the algorithm in this paper can gain better precision and practical effect for extracting buildings in remote sensing images,it has certain practicality and superiority.