Research On Aerial Image Mosaicking Based On High Mismatched Ground Object Recognition

Aerial images can be used to quickly obtain the surface semantic information and ground object locations,which become important visual geographic information products.Thus,aerial images are widely used in the fields of land and resources survey,urban planning,agricultural research,geological exploration,environmental change detection,disaster monitoring and rescue.When using aerial images,for the reason that a single aerial image contains limited surface area,multiple aerial images need to be spliced together.However,the same feature in the overlapping area of adjacent aerial images inevitably has differences in geometric position,texture and color.Therefore,a high-quality mosaic line needs to be selected during splicing to avoid large dislocation.The quality of mosaic line determines the quality of aerial image after mosaic to a certain extent.How to select a high-quality aerial image is significantly valuable.This study proposes a novel process for selecting aerial mosaic lines.Firstly,identify the high mismatched features in the overlapping area of aerial images,then construct a strategy to avoid these high mismatched features,and select the best pixels in the well matched area to form the best mosaic lines.The main results of this study are shown as follows:(1)Detect and locate building area in overlapping regions of aerial images.The Mask-RCNN neural network is used to fast detect the location of buildings in the overlapping regions and obtain their precise masks.After testing,the network constructed in this study can identify at least 87.53% of the building areas in the overlapping areas of aerial images,which achieves the detection and location of buildings in the overlapping areas of aerial images with high quality.(2)Position moving objects and evaluate the matching quality of corresponding pixels in the overlapping regions of aerial images.In this study,an exponential average cross-correlation algorithm is proposed to identify moving objects in the overlapping regions of aerial images and evaluate the matching degree of corresponding pixels.After testing,the method proposed in this study can detect and locate the moving objects in the overlapping regions of aerial images with high quality,and can clearly distinguish the matching degree difference of corresponding pixels in the overlapping regions of aerial images.(3)Quickly select the high-quality mosaic line of aerial images.This study proposes an optimized Dijkstra algorithm,which selects the optimal mosaic line pixel by pixel along the direction of the best matching of the corresponding pixels in the overlapping regions of aerial images.After testing,the mosaic line selected in this study can effectively avoid the high mismatched ground object area,and select the mosaic line along the well matched pixels to avoid visible dislocation.Then,this study proposes the mosaic line pyramid strategy,which greatly reduces the time for selecting aerial image mosaic line and improves the mosaic efficiency.By comparing this research method with three existing high-quality aerial image mosaic methods,results show that the method proposed in this study can obtain higher quality mosaic lines faster and achieve high-quality aerial image mosaic.