| Alone with the rapid development of urbanization, man-land relationship is thus an enormous issue facing China today, which has been put forwards stricter requirement to grasp the new patterns of urbanization scientifically. Urban change most commonly refers to building construction or demolition, and quickly extract the change information has important realistic significance for urban planning, illegal construction management and geographical database updating, etc. As the means and capability of very high resolution(VHR)satellite imagery acquisition become much easier, which has brought a new perspective and methodology to automatic building change detection. In fact, due to the dual influence of the VHR imagery spectral complexity and urban buildings itself exists huge disparity in tone, shape and size, much more theories and methods research will be necessary to automatic building change detection from VHR imagery.Feature changes will reflect the change of VHR image objects, and the stable and rich spatial features can describe the building’s attribute in different perspectives, so extract building change information by construct and fusion the difference of feature vectors set has full of theoretical feasibility. Therefore, in order to counter the shortage of multi-feature fusion approach for building change detection, morphological building index(MBI) and Dempster-shafer evidence theory(D-S) is introduced to explore the methodology of building feature expression and fusion. To achieve the goal of detecting building change information quickly and accurately, the main research content of this thesis includes:To begin with, in the original MBI algorithm, it doesn’t work well when buildings have relatively low contrast with the background. To deal with this problem, adaptive contrast enhancement algorithm was used to enhance the local contrast. Secondly, the original MBI algorithm cannot balance the various size difference of buildings, due to the fact that it only using one set of linear structure element. Therefore, a new approach contains three sizes linear structure elements:large, medium and small, which are extract MBI features separately and then calculate the maximum MBI value for each pixel.In addition, Using the improved MBI algorithm as the fundamental feature description for urban buildings, and combining the GLCM texture feature, phase congruency edge and pixel shape index. Then calculate each feature’s structural similarity as building change information source. It could reduce the false alarms ratio more effectively when compared with simply using MBI.Last but not least, by analyzing each feature’s resemblance-degree contribution to building change results, basic probability assignment function is constructed through introducing weighting factors, which can effectively integrate the advantages of each feature. And then it could decide whether the pixel is building change part based on D-S evidence combining theory.Two groups experiment with completely different building scenarios showed that correctness reached 93.26% and 93.77% respectively, the false alarms ratio is 3.3% and 1.76%, it gives a missed alarms ratio of 3.44% and 4.47% separately. Secondly, experiments with 3 groups VHR images which have different resolution level are all proceeded, and they both achieved satisfactory result except cause higher false alarms ratio when resolution increase to sub-meter. The experimental results indicate that the proposed method can be effectively applied to building change detection. |
PDF Full Text Request