Methods Of Extracting Aggregate Regions For SAR Image Semantic Segmentation

Synthetic Aperture Radar(SAR)has become an important method for earth observation which can be operated day and night under all weather conditions.Because of inherent imaging mechanism of SAR images,problems of speckle noise,shadow,and slope reduction on SAR images are caused,which make it difficult to interpret SAR images.Hierarchical visual semantic model for SAR images was proposed by our team,which uses semantic sketch segments as the basis of SAR Sketch Map,giving semantic information of sketch segment to guide segmentation and understanding of SAR images.The extraction of SAR image aggregate regions is one of the important steps from the primary semantic layer to the intermediate semantic layer in the hierarchical visual semantic model.In this paper,a geometric computational model based on semantic sketch segments and some related methods are established for problems existing in the extraction of SAR image aggregate regions.The main work of the dissertation includes:(1)Due to the problem of extracting the aggregate regions by using the circle operator algorithm is not only inaccurate,slow,and arcing at the boundary,this paper proposes an aggregate region extraction method based on single-edge aggregate segments of Sketch Map.In this method,an aggregate region extraction calculation model based on semantic sketch segments is established.Specifically,by using the spatial geometric position relationship of sketch segments and the topological relationship of the single-edge and double-edge aggregate segments,a geometric calculation model is established to extract SAR image's aggregate regions.The experimental results show that the proposed method is not only accurate for the extraction of the aggregate regions boundary but also has a significant improvement in the speed.(2)Considering that correct generation of aggregate segment set is premise of the extraction of aggregate regions,a model of generating aggregate segment set is proposed based on semantic information classification of sketch segment to solve problems in the process of generating aggregate segment set and unreasonable semantic information classification.In this method,semantic information of the sketch segment is clarified,and also the boundary sketch segment,independent target sketch segment,and aggregate region sketch segment are reasonably classified.Under the guidance of the determined semantic information, aggregate segment set is reasonably generated for extraction of aggregated region.The experimental results show that the proposed method improves the consistency of aggregate regions and pays more attention to the details of the regions.(3)For the problem of extracting the aggregate regions in large scene SAR images is very time-consuming,a method of extracting aggregate regions based on the block strategy is proposed on the basis of the aggregate region extraction calculation model.In this method,time complexity optimization is performed for each step in aggregate regions extraction process,including sketch map block acquisition,candidate aggregate region sketch segment and aggregate region block extraction.The experimental results show that the proposed method significantly reduces the time of extraction the aggregate regions in the large scene SAR image.