Research On Building Change Detection From High-resolution Remote Sensing Images In Complex Urban Scenes

In recent years,the continuous development of rapid acquisition and processing capacity for high-resolution remote sensing images and the prosperity of artificial intelligence have greatly promoted the automatic detection and identification of illegal or unauthorized buildings,and provided a systematic and efficient technique for urbanization monitoring and management.However,owing to the complexity of the space context in the urban scenaios,the otherness and diversity of the building structure,and the geometry offset and the position difference caused by relief displacements,artificial interpretation interpretation is still the main means in the practical application.In the process of this approach,the efficiency is very low,the waste of human and financial resources is serious,and a lot of manpower and material resources are required.Based on the above considerations,this paper takes the high-resolution remote sensing image in complex urban scenes as the starting point for building change detection,and explores relevant technical methods suitable for illegal or unauthorized building detection.The main research content of this paper includes the following three aspects:(1)Demolished building and changed rooftop detection method assisted with vector data.In the case of existing basic vector data,considering the mismatching between vector map and remote sensing images caused by relief displacement,this paper proposes a novel approach using graph-based model to locate rooftops and demolished buildings through the introduction of shape prior.The building region,boundary,and rooftop contour constraint terms were first formulated by multiple cues derived from both data sources.Next,roof-cut segmentation was performed by gathering all terms required for high-quality unsupervised rooftop extraction.Finally,the positional displacement statistics of similar adjacent buildings were collected to accurately estimate the rooftop location and achieve building demolition detection with the overlap ratio index.Owing to the appearance of illegal buildings with color steel shed on the roof,graph cuts based on superpixel for change detection on the roof is proposed.First,after obtaining the corresponding roof patches from bi-temporal images,object-oriented similarity measure is calculated by the SLIC algorithm and combining spectrum,texture and structure features.Then,the suspected changes on the roof are extracted with superpixel-based graph cuts and co-refinement.(2)New building extraction method assisted with vector data.On the basis of the above results,in order to detect new buildings,this paper proposes a novel two-step new building extraction method by the fully connected conditional random field.The hierarchical combinatorial grouping strategy and knowledge embedding are employed to effectively descibe the building target.A fully connected conditional random field model is introduced in this step to ensure that most of the buildings are incorporated.While it is hard to further remove the mislabled rooftops from the building candidates by only using classical features,we adopt stucture saliency cue as a new feature to determine whether there is a rooftop in each segmentation patch obtained from previous step.The saliency cue can provide an efficient probabilistic indication of the presence of rooftops,which helps to reduce false positives while without increasing false negatives at the same time.(3)Building change detection method from high-resolution remote sensing images.For building change detection between high-resolution remote sensing images,there are usually geometric deviations on the image and position differences between the images resulting from relief displacement.In order to obtain the results with high precision and strong reliability,a novel patch-based matching approach is developed using densely connected conditional random field(CRF)optimization to detect building changes from bi-temporal aerial images in this paper.First,the bi-temporal aerial images are combined to obtain change information using an object-oriented technique,and then semantic segmentation based on a deep convolutional neural network is used to extract building areas.With the change information and extracted buildings,a graph-cuts-based segmentation algorithm is applied to generate the bi-temporal changed building proposals.Next,in the bi-temporal changed building proposals,corner and edge information are integrated for feature detection through a phase congruency(PC)model,and the structural feature descriptor,called the histogram of orientated PC,is used to perform patch-based roof matching.We determined the final change in buildings by gathering matched roof and bi-temporal changed building proposals using co-refinement based on CRF,which were further classified as “newly built,” “demolished”,or “changed”.