Abstract:Wide-Field Small Aperture Telescopes (WFSAT) are widely used for surveilling space objects. Due to their wide-field of view (FOV) characteristics, these telescopes can cover a large areas of the sky at once, improving observation efficiency. However, a wide-field optical telescope is highly sensitive to external stray light (such as moonlight and thin clouds), which can significantly reduce the quality of observation data. In severe cases, it can cause the telescope to malfunction and inaccurately position the object. In response to this problem, this paper proposes a model for suppressing stray light in astronomical images based on deep learning: the Pyramid Deformable Large Kernel Attention (PD-LKA) Model. This model expands the receptive field through a pyramid structure, captures multi-scale features, and improves the model's robustness to various scales of stray light interference. Meanwhile, through the Deformable Large Kernel Attention (D-LKA), the model can more accurately locate and enhance the feature extraction ability in areas affected by stray light interference, thereby better suppressing stray light.Using simulated astronomical image pairs to train the model, the tests achieved a PSNR of up to 32.540 and an SSIM of up to 0.938. Finally, the model is applied to a image sequence with real stray light interference. The restored images undergo astronomical positioning and orbital association processing. The results show that the positioning accuracy of the object is better than 5 arcseconds. This indicates that the model proposed in this paper not only recovers the object and background stars but also effectively preserves their gray values, shapes, and positional information.

The Deep Neural Network Based Photometry and Astrometry Framework for Wide Field Small Aperture Telescopes

Detection and Classification of Astronomical Targets with Deep Neural Networks in Wide Field Small Aperture Telescopes

Point spread function estimation for wide field small aperture telescopes with deep neural networks and calibration data

Effective Framework for Space Target Detection Through Atmospheric Turbulence

Deep Neural Network closed-loop with raw data for optical resident space object detection

CSST Dense Star Field Preparation: A Framework for Astrometry and Photometry for Dense Star Field Images Obtained by the China Space Station Telescope (CSST)

Blind Deconvolution with Principal Components Analysis for Wide-Field and Small-Aperture Telescopes

Neural-network-based wavefront solution algorithm for a wide field survey telescope

Point Spread Function Modelling for Wide Field Small Aperture Telescopes with a Denoising Autoencoder

Deep Learning-Based Small Target Detection for Satellite–Ground Free Space Optical Communications

Neural Network Based Point Spread Function Deconvolution For Astronomical Applications

Perception of Misalignment States for Sky Survey Telescopes with the Digital Twin and the Deep Neural Networks

A model for suppressing stray light in astronomical images based on deep learning

Neural Network Assisted High Flexibility and High Resolution Shack-Hartmann Wavefront Sensing for Astronomical Observation in Darker Sky Areas

Fast Inspection and Accurate Recognition of Target Objects for Astronaut Robots Through Deep Learning

Optical Transient Object Classification in Wide Field Small Aperture Telescopes with Neural Networks

Deep learning-assisted near-Earth asteroid tracking in astronomical images

Machine Learning for Improving Stellar Image-based Alignment in Wide-field Telescopes

Application of Deep Learning Methods Combined with Physical Background in Wide Field of View Imaging Atmospheric Cherenkov Telescopes

Deep learning solutions to telescope pointing and guiding

PI-AstroDeconv: A Physics-Informed Unsupervised Learning Method for Astronomical Image Deconvolution