CSST-PSFNet: A Point Spread Function Reconstruction Model for the CSST Based on Deep Learning

TL;DR

Results indicate that CSST-PSFNet provides a flexible and extensible framework for future on-orbit PSF calibration in large-scale CSST surveys, with potential applications in weak-lensing cosmology and precision astrophysical measurements.

Abstract

This paper presents CSST-PSFNet, a deep learning method for high-fidelity point spread function (PSF) reconstruction developed for the Chinese Space Station Survey Telescope (CSST). The model integrates a residual neural network, a lightweight Transformer architecture, and a variational latent representation to address key challenges in CSST imaging, including severe PSF undersampling, inter-band variability, and smooth spatial variation across the focal plane. Trained and validated on high-resolution star-PSF pairs generated by the CSST Main Survey Simulator, CSST-PSFNet achieves improved pixel-level accuracy and more precise recovery of shape parameters relevant to weak lensing compared to widely used PSFEx. On both the standard test dataset and a blurred dataset representing the upper bound of expected on-orbit PSF degradation, the model achieves a size residual precision below 0.005 and an ellipticity residual precision below 0.002. A weak-label adaptation experiment further shows that the model can recover PSFEx-level performance when the true PSF is unknown, demonstrating robustness in controlled degradation scenarios and weak-label adaptation experiments. These results indicate that CSST-PSFNet provides a flexible and extensible framework for future on-orbit PSF calibration in large-scale CSST surveys, with potential applications in weak-lensing cosmology and precision astrophysical measurements.

Figures (10)

• Figure 1: Representative PSFs at the detector center from all 18 photometric CCDs in DATASET 2. For each CCD, the PSF is extracted from the ground-truth spatial PSF field of DATASET 2 at the grid position closest to the CCD center. The labels above each subfigure indicate the CCD index, photometric band, effective wavelength, FWHM, and PSF ellipticity $e$. The full spatially varying PSF fields in DATASET 2 are used as the ground truth for evaluating the corresponding PSF reconstruction results. The color scale represents the normalized PSF intensity and is displayed on a logarithmic scale with logarithmically spaced color-bar ticks, highlighting both the PSF core and low-intensity wing structures. Note that the displayed PSFs are shown on the $2\times$ upsampled grid (0.037″/pixel) for visualization, but the quoted FWHM values are reported in native CSST pixels (0.074″/pixel) for consistency with instrument specifications; equivalently, the FWHM on the displayed grid is twice the quoted value.
• Figure 2: Same as Figure \ref{['fig:ground_truth_test_PSF']}, but for DATASET 3. Representative PSFs at the detector center are extracted from the ground-truth spatial PSF fields of DATASET 3 and used to assess the PSF reconstruction performance under this dataset. The color scale represents the normalized PSF intensity and is shown on a logarithmic scale with logarithmically spaced color-bar ticks, enabling visualization of both the PSF core and faint wing structures.
• Figure 3: Architecture of CSST-PSFNet, integrating convolutional encoding, CCD/position conditioning, and Transformer-based dual-query decoding for PSF reconstruction. Bottom: module details of ResBlock, Pos Block, CCD embed/FiLM, and UpsampleBlock.
• Figure 4: Residual maps of the central PSFs reconstructed by CSST-PSFNet for all 18 CCDs on DATASET 2. Compared to the PSFEx results (Figure \ref{['fig:psfex_residual']}), the model residuals are typically an order of magnitude smaller in RMS ($\sim10^{-5}$ versus $\sim10^{-4}$), and large-scale ring-like artifacts are absent. Only fine high-frequency structures of low amplitude remain, demonstrating the model’s improved ability to capture both PSF cores and wings across the full focal plane.
• Figure 5: Residual maps of the central PSFs reconstructed by PSFEx for all 18 CCDs on DATASET 2, relative to the ground truth PSFs shown in Figure \ref{['fig:ground_truth_test_PSF']}. The reconstructions reproduce the overall morphology but leave prominent features, most notably bright central cores (red) surrounded by negative rings (blue). These features indicate difficulties in simultaneously matching both the PSF core and extended wings.