极 端 太 空 成 像 环 境 下 的 航 天 器 光 学 图 像仿 真 数 据 集 构 建

Objective Optical imaging is one of the primary means for detecting and identifying space targets such as spacecraft. It provides critical information for analyzing spacecraft flight attitude and operational status, and holds significant value for on-orbit condition monitoring and fault localization. However, optical imaging of spacecraft can be degraded by the extreme space environment, resulting in the loss of spacecraft characteristic information and posing great difficulties for structural analysis and target recognition. Therefore, in-depth research on algorithms for spacecraft image restoration and component detection and recognition is required. Since real spacecraft image data acquired by actual imaging observation equipment are generally not publicly available, researchers face difficulties in obtaining spacecraft optical imaging data under real mission scenarios. Consequently, synthetic image data generation has become an important approach to addressing the data scarcity issue. Hence, it is necessary to construct a dataset of spacecraft optical images under extreme space imaging conditions that meets the requirements of realistic scene simulation and fully accounts for various space degradation effects. Methods In this study, the imaging characteristics and degradation effects of the extreme space environment were first analyzed. Based on this analysis, a workflow for constructing a simulated spacecraft optical image dataset was specifically designed. Detailed methods were presented for the analysis and modeling of the spacecraft optical imaging process, target simulation and image generation, blur degradation simulation, and halo artifact simulation. The performance of spacecraft image simulation under different intensities, illumination and observation conditions, and varying degrees of degradation effects was analyzed, thereby validating the accuracy of the proposed method for constructing a spacecraft optical image simulation dataset. Results and Discussions The dataset contained optical simulation images of four typical spacecraft models. Based on 200 baseline images generated for each spacecraft model under different viewpoint-illumination combinations, seven additional imaging conditions were simulated for each corresponding viewpoint-illumination scenario: bright target, dim target, mild blur, severe blur, mild halo artifact, severe halo artifact, and blur combined with halo artifact. Thus, each spacecraft model had 1600 multi-view images under eight imaging conditions. The total number of simulated images for the four spacecraft targets amounted to 6400. The dataset could be further enriched by increasing the number of target models and imaging conditions according to the proposed method, thereby continuously improving the quantity of images in the dataset. The dataset accounted for extremely dark imaging backgrounds, blur effects caused by atmospheric scattering, and halo artifacts due to overexposure of bright targets, thus capturing the main features of extreme conditions in spacecraft imaging. Furthermore, based on distinguishing various imaging degradation conditions, each image was annotated using LabelMe software for spacecraft components, which enabled object part bounding box detection and pixel-level segmentation under different degradation types. The generated JavaScript Object Notation (JSON) annotation files included pixel-level positions and labels of spacecraft component regions. Consistency between ten pairs of simulated and measured spacecraft images was evaluated using a seven-point Likert scale. All ten image pairs received scores above 5 (indicating relatively high consistency), with an average score of 6.52, reaching the 'very consistent'criterion. This demonstrated that the simulated spacecraft optical images generated by the proposed method were highly consistent with measured images. Conclusions To meet the application requirements for training and testing spacecraft optical image processing and component recognition algorithms, and to satisfy the demands of realistic spacecraft optical imaging scene simulation, this study investigates methods for constructing a simulated spacecraft optical image dataset under extreme space imaging environments. First, the imaging characteristics of extreme space environments, such as extremely dark backgrounds, directional light sources, and atmospheric effects on ground-based imaging, are analyzed, along with imaging degradation effects including dim targets, blur degradation, extreme brightness contrast, and overexposure. Based on this analysis, a workflow for constructing the simulated spacecraft optical image dataset is specifically designed. Detailed methods for imaging process analysis and modeling, target simulation and image generation, blur degradation simulation, and halo artifact simulation are presented. The performance of spacecraft optical image simulation under different intensities, illumination and observation conditions, and varying degradation effects is evaluated. The results indicate that the proposed method for constructing a spacecraft optical image dataset can generate simulated spacecraft optical images under various imaging conditions. The image dataset constructed using this method can effectively support the training of spacecraft image recognition algorithms under different observation scenarios and environmental conditions.