Research On Small Body Observing Regulation Of 3D Modeling

Nowadays,small body exploration has become an important project in deep space exploration,which is of great significance to the development of space economy and the verification of deep space exploration technology.The three-dimensional mapping of small bodys is an important part of small body exploration.The threedimensional surveying and mapping work aims at taking images of small celestial bodies,transmitting them back to the ground,and establishing the three-dimensional model through the modeling program.Because the high-definition image can only cover a very small part of the surface of small celestial bodies,in order to establish a complete three-dimensional model of small celestial bodies,the detector needs to take a large number of high-definition images,which creates pressure on the onboard storage and communication links of the detector.Therefore,it is a challenge to plan the shooting work in advance and save the shooting resources to the maximum extent on the premise of ensuring the correct operation of the modeling program.This paper is the first exploration of machine learning method in aerospace mission.This paper first introduces the process of small body exploration,and then puts forward the mapping planning technical route of "planning,photographing,modeling,re planning and re photographing" from far and near.In the process of approaching to the small body,the line of sight direction of Surveying and mapping observation in the next stage is planned by using the rough model constructed in the past.In this paper,the reward function of strengthening learning in small celestial mapping planning is proposed,and the problem of mapping planning is solved by using the method of strengthening learning.This paper mainly includes:1)Aiming at the method of 3D reconstruction of small bodys based on photometry(SPC),this paper analyzes the requirements of the algorithm for camera pointing,designs a reward function,abstracts the shooting position and shooting direction of the detector as action,and finally extracts the p ast shooting history as state.According to the actual situation of the surveying and mapping process of small body exploration,the elements of Surveying and mapping observation planning task are mapped to the basic elements of reinforcement learning,and the framework of reinforcement learning of Surveying and mapping observation is established.Prepare for the combination of planning tasks and reinforcement learning.2)In view of the problem that the state action space dimension is too large and the deep neural network used for decision-making is too complex in the intensive learning of small body mapping planning,the neural network optimization method of understanding coupling is proposed to simplify the network complexity and accelerate the convergence speed of training.The method proposed in this paper effectively improves the observation efficiency in the process of Surveying and mapping small celestial bodies,a nd reduces the amount of data transmission between heaven and earth.