Man-made Object Detection For Remote Sensing Imagery Via Graph Model And Manifold Ranking

With the constant improvement of satellite technology,remote sensing images have offered a wide range of data sources for the acquisition of spatial information.The detection of man-made object,such as buildings,roads and bridges,is one of the fundamental but challenging tasks in image interpretation and target recognition.In recent years,automatic detection of man-made objects from remote sensing images has been applied in many applications,including urban planning,disaster evaluation,change detection and map updating.However,due to the complexity of the scene,it is still difficult to recognize targets completely.Based on high resolution remote sensing images,we measure the detection of the man-made objects as a ranking problem on the manifold structure,and put forward an algorithm for extracting man-made objects automatically by integrating a graph model with the manifold ranking algorithm.Initially,we estimate a priori value of the man-made objects with the use of symmetric and contrast features.Quantum genetic algorithm is combined with the KSW entropy method to enhance computational efficiency of the contrast prior detection.According to the property that the appearance of man-made objects often present symmetrical features,superpixels are adopted as deformable maximal disc hypotheses,so we can obtain the symmetry prior by machine learning.After computing a priori value of symmetry and contrast,we obtain the final priori of man-made object for optimization.Then,we build a graph model to simulate the manifold structure of data,and use superpixels as the nodes of graph to reduce the computation complexity.We convert the target detection into an optimal cutting problem.Multiple characteristics,namely colour,texture and main direction,are used to compute the similarity of the adjacent nodes.In this way,we give full consideration to the local spatial information in the image.Finally,manifold ranking algorithm are used to optimize the priori value of man-made objects.Based on the graph model,the priori value are labeled as known nodes.The probability of each node being a part of a man-made object is computed on the basis of its relevance to the known nodes.The man-made objects are then segmented according to the ranking map,and we can obtain the detection result of man-made objects from remote sensing images automatically.Graph model and manifold ranking algorithm which are chosen to optimize priori value give full consideration of adjacency relation in remote sensing image,using adjacent nodes in graph to transfer the category value.Due to the spatial correlation in the distribution of man-made object is obvious,we make full use of this feature to improve the accuracy,and make the final detection result more reliable.We have evaluated the proposed algorithm on 2 aerial images and 60 images from UC Merced Land Use Dataset.The experimental results show that our algorithm has the applicability in man-made object detection,and can extract targets with high recognition rate and low omission rate.Compared to the previous algorithms,the proposed approach has presented decent overall quality.