Research On Object-oriented Building Extraction And Change Detection Technology Of High Resolution Remote Sensing Imagery

As a kind of important infrastructure in geodatabase,the automatic extraction and change detection of building has been a focused and difficult research problem in the field of artificial intelligence and photogrammetry and remote sensing.With the ability of directly and quickly obtaining large-scale region highly recognizable imagery,high resolution remote sensing imagery can provide reliable data resources for building extraction and building change detection.However,due to the factors such as details are too rich in high resolution remote sensing imagery and complex building structure in real scene,automatic building extraction and building change detection based on high resolution remote sensing imagery is still a challenging task.Under this background,the thesis focuses on the method of building extraction and building change detection based on high resolution remote sensing imagery.The major work and innovations of the dissertation are listed as follow:1.After introduction the background and significance of building extraction and building change detection,methods of shadow detection,building detection and building change detection are concluded,commonly used building structure features are summarized,and object-oriented imagery analysis is introduced.2.Method of shadow detection based on multi-scale segmentation and morphology operation is proposed.The proposed method adopts the technical route of "pixel-object-pixel".First,object-based shadow index is obtained by objected morphology dilation and erosion operation after imagery segmentation.Then,shadow detection result is obtained by shadow index vector and multi-scale brightness mean,which are constructed by multi-scale segmentation.The experimental results verify the effectiveness of the proposed method.3.Aiming at the building extraction from high resolution remote sensing imagery,a building extraction method using multi-feature is proposed.The proposed method is based on spectral,gradient,spatial,contextual and textural features and combines multi-scale imagery segmentation result.First,based on imagery processing by bilateral filtering,the building index is constructed using multi-scale and multi-direction gradient operators,and some rectangle buildings are extracted by the building index,the morphology operation and the shape features.Subsequently,the voting matrix is calculated by shadow features and eight-direction linear structural elements.And the recognition of the building samples is completed by the shadow features and light direction,which is determined by voting matrix.Next,the features(spectral and textural features)of the building samples are spatially clustered,and then Gaussian model was used to perform probabilistic modeling to complete pixel-level building extraction.Finally,the final building extraction results are achieved with the combination of the pixel-level results and multi-scale segmentation.Experimental results show that the proposed method can effectively extract building objects with high stability and great applicability,which is able to provide reliable information for building change detection.4.A building change detection method based on fuzzy fusion is proposed.The strategy of building extraction and change detection at the same time is designed in the proposed.First,objects are generated from the result of imagery fusion by imagery segmentation,and the feature distances of the object including texture,spectrum and edge are calculated.Then,the total factor change detection can be completed by fusing these feature distances based on the fuzzy set theory.Finally,through the spatial location superposition method,building change is realized based on change detection and building extraction.The experimental results show that the proposed method can combine the advantages of each feature with high accuracy and great applicability.