Digital Surface Model Generation Of Weak Texture Regions In Optical Satellite Remote Sensing Images

Digital Surface Model(DSM)is a discrete expression of earth’s surface information.It can provide elevation of specific areas or targets and is used for multiple tasks such as mapping,smart city and battlefield awareness.For decades,DSM generation is always one of the most important task of the surveying and mapping department.Currently,respect to space stereo mapping,there is no mature interferometric radar satellite in China.But an optical stereo mapping satellite constellation has been built,which contains four satellites,ZY301,ZY302,ZY303 and GF7,and can meet the need of stereo mapping scale from 1:50000 to 1:10000.Under this background,optical satellite remote sensing image dense matching will be a main way of DSM generation for the surveying and mapping department in our country.Although,dense matching can extract DSM with height accuracy from optical images in most cases,it still has difficulties with weak texture,disparity discontinuities and occlusion and.Manual editing is often requited for the extracted DSM from these areas before real applications.The conflict between the low precision of DSM extracted by dense matching in some areas and the high requirement of various applications will be more severe with the continuous improvement of image resolution.For the purpose of improving the quality and automation of DSM,this paper studies dense matching theory and difficulties mentioned,and proposes improvement strategies.The main work as follows:1.Dense matching theory is studied and the difficulties in it are analyzed.The dense matching problem is defined,and the(semi)global solution mathematical model is constructed.Then,the key factors in the model are analyzed and the optimal criterions are given.On this basis,the reason for the difficulty of SGM in the disparity discontinuities and weak texture area is deeply explored,and the improvement direction is pointed out.Aiming at the problem of large amount of computational redundancy in SGM cost aggregation,inspired by the shortest path strategy,the single path cost aggregation is regarded as the determination of the minimum cost path,and the disparity selection tendency between adjacent pixels based on smoothing constraint is changed from aggregation cost to candidate disparity value,which makes cost aggregation independent of disparity range and greatly improves processing efficiency.2.The DSM extraction method of calm water regions based on plane constraint is studied.Aiming at the difficulty in water matching,a water extraticon strategy based on seed point detection and growth is proposed firstly.Then,an adaptive match primitive selection strategy that takes into account geometric deformation is proposed,which overcomes the problem of unreliable matching cost caused by the lack of gray-level change information in the traditional region matching primitives in water regions.Combining the characteristics of the optical remote sensing satellite stereo image pair,a robust matching cost is designed.The reliable water region match result is obtained after mismatch elimination and interpolation.Finally,based on the assumption that the elevations are approximately consistent in the same calm water region,a DSM generation strategy based on planeconstraint is proposed,which realizes the reliable extraction of DSM in calm water regions.3.The DSM extraction method of buildings based on the shading and line feature constraints is studied.Aiming at the difficulty in buildings matching,the disparity oversmoothing problem of SGM in this scene is divided into two types: the disparity estimation value is too large and the disparity estimation value is too small.The shadow and line feature information are introduced to solve the two problems.A shadow extraction strategy based on histogram statistics and a strategy for determining the relationship between shadows and buildings based on the disparity map and RPC are proposed to obtain reliable disparity estimates in the shadow area,which improves the problem of large disparity estimates in this area.Based on dyadic wavelet edge detection and direction-constrained edge tracking,reliable line features are extracted,and then a multi-constrained line feature matching algorithm fusing structure and gray information is proposed to achieve reliable matching of line features.Finally,by combining the shadow and disparity map to determine the line feature attribution,and performing the line feature pairing process,adaptively determine the line feature refined disparity map method,which improves the problem of the small disparity estimation value of the building boundary area.In summary,aiming at the matching problems of calm water scenes and building scenes in DSM generation from optical satellite remote sensing images,this paper proposes a fast cost aggregation method based on minimum cost path,a calm water DSM generation method based on plane constraints and a building DSM generation method based on shadow and line constraints.ZY3 and GF7 satellite images,covered different calm water scenes and building scenes,are used to fully verify the effectiveness and applicability of the proposed methods.Through a comparative analysis with the classic SGM,industry-leading commercial software Geomatica and checkpoints,it shows that the proposed methods can improve the efficiency of dense matching and the DSM quality of calm water scenes and building scenes.