Research On Visual Image Algorithm For On-orbit Awareness Of Micro-satellites

With the maturity of microsatellite technology,the number of launches of microsatellites continues to increase,and its application scenarios are constantly expanding.Recently,on-orbit awareness gradually becomes a research hotspot in the aerospace field.Image processing algorithm based on visible light vision is a key of on-orbit sensing.Due to the complicated environment in space and the limited computing and storage resources of microsatellite,the related image processing algorithms are facing new problems and challenges.This paper studies the image processing algorithms involved in the on-orbit sensing link in the above application scenario.For long-distance sensing,the stray light suppression algorithm of space background image and small object detection are studied in this paper.For close-range sensing,the detection and recognition of components on spacecraft is studied.Firstly the stray light suppression algorithm for space background images is studied.In order to restrain stray light on image,a full convolutional neural network is designed,which is based on the analysis of image noise characteristics.The network is trained on a simulation images dataset.An experiment is carried out based on the images taken on-orbit by TT2-Sat.The results verify the effectiveness of the algorithm in removing stray light.Secondly the small object detection algorithm is studied for space-based target observation task.An improved MTI algorithm is proposed in this paper to solve the problem of space objects detection in images taken by microsatellite.In order to detect the inconsecutive target's trajectory,at the beginning of the algorithm we set a special preprocessing which is called pixel's feeling domain.To reduce the time of the algorithm,we simplified the time projection part of the classic MTI algorithm.Finally,targets trajectory are obtained through connected domain detection.The experiments on TT2-Sat's images show that,the improved MTI algorithm can effectively eliminate the background and is suitable for the inconsecutive target's trajectory detection.In addition,the algorithm's processing speed almost meets the real-time task.Thirdly the algorithm of detection and recognition of components on spacecraft is studied.The Faster RCNN algorithm is applied to detect and classify the satellite's components.The base network of Faster RCNN is replaced with MobileNet in this paper,in order to deploy the algorithm on microsatellite.An image dataset of TT3-Sat's components is proposed.The MobileNet-Faster RCNN is finetuned on TT3 dataset.The experiment shows that the network model in this paper can accurately detect and identify camera and antenna on TT3-Sat.