The acquisition and application of spacecraft optical data is an important part of space-based situational awareness(SSA).Spacecraft optical data processing techniques can assist in tasks such as on-orbit operation,sp...The acquisition and application of spacecraft optical data is an important part of space-based situational awareness(SSA).Spacecraft optical data processing techniques can assist in tasks such as on-orbit operation,space debris removal,and deep space exploration.However,the extreme lack of real spacecraft optical data is an insurmountable difficulty,which hinders the development of deep learning-based data processing techniques.Existing synthetic datasets usually only contain visible-light images,only support a specific task,and lack diversity in the scale of the spacecraft,which cannot adapt to actual application environments.Therefore,we propose a multi-modal,multi-task,and multi-scale spacecraft optical dataset(TriM-SOD),which has 3 superiorities:(a)multi-modal:it includes data in various modals,such as visible light and infrared;(b)multi-task:it includes labels for multiple tasks,such as spacecraft detection and spacecraft component segmentation;and(c)multi-scale:it features a variety of sizes for spacecraft in the images.To validate the effectiveness of our dataset and evaluate the performance of methods in the tasks,we use TriM-SOD to train and test several typical or recent methods for object detection and semantic segmentation.TriM-SOD has been made public and can be used as a benchmark to further promote the future development of SSA.展开更多
基金supported by the National Natural Science Foundation of China(62331006,62171038,62088101)the Fundamental Research Funds for the Central Universities.
文摘The acquisition and application of spacecraft optical data is an important part of space-based situational awareness(SSA).Spacecraft optical data processing techniques can assist in tasks such as on-orbit operation,space debris removal,and deep space exploration.However,the extreme lack of real spacecraft optical data is an insurmountable difficulty,which hinders the development of deep learning-based data processing techniques.Existing synthetic datasets usually only contain visible-light images,only support a specific task,and lack diversity in the scale of the spacecraft,which cannot adapt to actual application environments.Therefore,we propose a multi-modal,multi-task,and multi-scale spacecraft optical dataset(TriM-SOD),which has 3 superiorities:(a)multi-modal:it includes data in various modals,such as visible light and infrared;(b)multi-task:it includes labels for multiple tasks,such as spacecraft detection and spacecraft component segmentation;and(c)multi-scale:it features a variety of sizes for spacecraft in the images.To validate the effectiveness of our dataset and evaluate the performance of methods in the tasks,we use TriM-SOD to train and test several typical or recent methods for object detection and semantic segmentation.TriM-SOD has been made public and can be used as a benchmark to further promote the future development of SSA.
