Dynamic Parameter Identification Method for Satellite Flexible Solar Array Considering Illumination Characteristics

Non-contact visual measurement methods are widely employed for the measurement and identification of onorbit dynamic parameters of satellite flexible solar arrays. During prolonged satellite operation，onboard cameras generate substantial image data，which are frequently compromised by lighting-induced failures such as over-exposure and under-exposure. To address the challenges of eliminating degraded images and performing parameter identification on valid data， a multi-depth parallel convolution module is developed. This module preserves feature information from both early and late convolutional stages，enabling synchronous multi-dimensional feature extraction. An ultra-lightweight convolutional neural network is constructed based on this module，significantly reducing parameter complexity. The network is trained using calibrated datasets，validated through test datasets，and its output images are utilized for spacecraft dynamic parameter identification. Experimental results demonstrate that this lightweight architecture achieves 99. 60% identification accuracy with fewer than 0. 1 million parameters. The processed images enable accurate calculation of solar array dynamic parameters under complex orbital conditions.