Multi-pattern 3D Intelligent Reconstruction Method For Non-cooperative Space Targets Based On Deep Learning

The space target 3D reconstruction technology is one of the key technologies in on-orbit service.It can provide high-precision navigation models for on-orbit clearance and rescue missions for space non-cooperative targets such as failed satellites and uncontrolled spacecraft.Due to the hi gh complexity of the space environment,only the use of visible light images to achieve the 3D reconstruction of the target is facing challenges.Therefore,this paper proposes a spatial target 3D reconstruction method based on visible light image reconstruction point cloud and lidar point cloud intelligent fusion.The contents of the study are as follows.First,we study a 3D reconstruction method based on lidar for spatial non-cooperative targets.In a complex spatial light environment,Lidar has certain advantages.Based on the simulation of the working principle of laser radar,the point cloud data acquisit ion,point cloud registration,three-dimensional grid structure and texture mapping were studied.The three-dimensional reconstruction of the target model was realized,and the three-dimensional reconstruction based on lidar was simulated.Secondly,the method for analyzing the visual quality parameters of space targets is explored,and the image quality,texture richness,and exposure are used to influence the quality parameters of the 3D reconstruction of vis ible light images.Mathematical models of image quality are established,and three qualities are used by BP neural network.The parameters are trained to achieve intelligent optimization of the input image sequence.Third,based on the CNN neural network caffe open platform to build the deep learning framework of Faster RCNN,construct the feature extraction network and RPN structure framework,realize the intelligent identification of the target module in the image sequence,and then filter out with certain components and clear Complete target image.Finally,a mult i-modal 3D reconstruction method based on visible light image reconstruction point cloud and Lidar scanning point cloud is proposed,and the true point recovery and target integrity restoration of the reconstruction point cloud are achieved by using the nearest point iteration algorithm of curvature.In this dissertation,a large number of 3D reconstruction experiments were performed on the simulation model and compared with the origina l model to perform error analysis to verify the validity of the above research contents.