Research On Region Map Acquisition Method In High Resolution Large Scene SAR Image Segmentation

Synthetic Aperture Radar(SAR)can acquire high-resolution images all day and around the clock,which is very important for Earth observation.Due to the inherent imaging mechanism of SAR images,the problem of speckle noise and high-dimensional heterogeneity on SAR images is very difficult,which makes the interpretation of SAR images very difficult.The SAR image hierarchical visual semantic model proposed by the team uses the sketch line segment as the base element to guide the SAR image interpretation by giving the sketch line segment semantic information,and transforms from pixel space to semantic space,which improves the segmentation effect to some extent.As an intermediate semantic layer of hierarchical visual semantic model,SAR image area map is a key step in SAR image interpretation.Based on the Sketch Map,this paper proposes a method for obtaining regional maps based on spatial constraints and KD(K-Dimensional)trees for the large-scale SAR image features and the large number of sketch segments in high resolution large scenes.The main work includes:(1)Give the sketch line segment clear semantic information.There are many sketches of SAR images in large scenes.There are many semantic possibilities in the process of generating sets of clustered segments.The ambiguous semantics have great influence on the extraction of region graphs.This paper proposes a method for assigning semantic information of sketch segments based on spatial constraints.Through the spatial constraint,the semantic information of the line segment is clarified,and the semantic information of each sketch line segment is unilaterally aggregated,double-sided aggregated,end-side horizontal cascading,and surrounding line segments are given.The experimental results show that the proposed method can significantly improve the semantic information of semantic line segments.(2)According to the semantic information of the line segment,the KD tree-based aggregation set calculation method is proposed.The large-scale SAR image sketch line segment is more.The 3543×1506 size SAR image will still obtain 9075 sketch line segments under certain sparse degree,based on the table method.The aggregate set calculation method cannot save the k-nearest neighbor relationship of the semantic line segment because it adopts the linear structure organization element.When calculating the neighbor,it involves a large number of neighbor search operations,and the resource is increased in the case of the large number of semantic segments of large-scale SAR images.The consumption is serious.Therefore,this paper proposes a KD tree-based aggregation set calculation method.KD tree can effectively preserve the spatial neighbor relationship due to its structural characteristics,which greatly improves the efficiency of algorithm search,not only makes the semantic line segment set calculation effect,but also has a higher speed improvement.(3)Based on the obtained aggregated set,a large-scale SAR image region map extraction method based on spatial constraints and KD tree is proposed.Firstly,the proposed block fusion strategy is used to obtain the Sketch Map of SAR images.Then,for the aggregation area of large-scale SAR images,the problem of severe resource consumption is solved.Based on the obtained set of aggregated line segments,the proposed KD-based method is used.The SAR image gathers the region result.Then the structure region is obtained by using the proposed geometric structure window and the neighboring connection to surround the closed method.And finally the respective regions obtained by the fusion obtain the final region map.The experimental results show that for large-scale SAR images in high-resolution large scenes of 3543×1506 size,this method can obtain the area map of the original image efficiently and accurately.