Research On The Multi-source Data Fusion Algorithm Of Ground-based SAR Based On Generative Adversarial Networks

Ground based synthetic aperture radar(Ground-based SAR)system has the technical advantages of non-contact,high-precision and large area continuous monitoring.Besides,it is an important method for regional monitoring,surface deformation monitoring and fixed-point continuous measurement.The traditional multi-source data fusion algorithm of Ground-based SAR image is usually based on coordinate transformation or feature matching.It needs to manually extract the coordinates of Ground-based SAR system and monitoring points for conversion calculation,which is difficult to realize automatization.The multi-source data fusion of Ground-based SAR image can also be realized by using the generated countermeasure network model.However,due to the unique imaging mechanism of Ground-based SAR,the correlation between Ground-based SAR image and optical image is low,so it can not be trained directly.In order to solve this problem,based on the grid projection method,this paper completes the coordinate unification of 3D point cloud terrain data and Ground-based SAR image,realizes the spatial fusion of multisource data,generates the Ground-based SAR image in the east-north coordinate system,and takes it as the training sample of cyclic consistent generation countermeasure network to carry out the research of Ground-based SAR multi-source data fusion algorithm based on generation countermeasure network.The main research work is as follows:(1)Because the coordinate systems of Ground-based SAR image and optical image are not consistent,the ground features of Ground-based SAR image and optical image can not be directly established corresponding relationship,they can not be directly trained as samples.In order to solve these problems,in this paper,a fusion method of Ground-based SAR image and 3D point cloud data based on raster projection is proposed.Firstly,the transformation relationship between the range azimuth coordinate system and the northeast sky coordinate system is derived under the condition of arbitrary Ground-based SAR observation geometry.Then,through the grid projection method,the corresponding relationship matrix of Ground-based SAR image information height information in the east-north coordinate system is established,and the ground-based SAR image and 3D point cloud data are fused to generate the Groundbased SAR image in the east-north coordinate system,and complete the direct quantitative analysis of the coordinate conversion accuracy.The experimental results show that the Ground-based SAR image is accurately transformed into the east-north coordinate system and can be used as the training sample set.(2)In order to solve the problem of pitch angle ambiguity in the fusion of 3D point cloud data and Ground-based SAR in target monitoring of complex structure,which produces invalid features and affects the quality of training samples,this paper analyzes the reasons for pitch angle ambiguity in the process of target monitoring of complex structure,and proposes a pitch angle ambiguity processing method for the fusion of 3D point cloud data and Ground-based SAR based on grid projection.The experimental results show that the effect of false image removal in training samples is obvious,which shows that this method has good processing effect on pitch angle ambiguity.(3)In order to solve the problem that the existing fusion methods of Ground-based SAR image and optical image are difficult to distinguish the features of the whole monitoring scene,a multi-source data fusion algorithm of Ground-based SAR based on Generative Adversarial Networks is proposed.Using the Ground-based SAR image and Google optical image in the east-north coordinate system as training samples,unsupervised learning is carried out to train the mapping relationship between the Ground-based SAR image pixels and Google optical image pixels in the east-north coordinate system to realize the data fusion of 3D point cloud data,optical image and ground-based SAR image.The experimental results show that this method can output realistic optical images of the same scene according to the east-north coordinate system,and complete multi-source data fusion.