The AI supervisor of source-extraction algorithms for images obtained by wide-field small-aperture optical telescopes

Aims . Wide-field small-aperture optical telescopes are essential for the imaging of celestial objects for time-domain astronomy. The extraction positions and magnitudes of celestial objects within observation images are a key prerequisite for carrying out further scientific results. The parameters of the source-extraction algorithms must be fine-tuned to achieve an optimal performance. This can be time-consuming and resource intensive. Methods . Inspired by the manual parameter fine-tuning procedure, we propose the concept of an AI supervisor for source-extraction algorithms based on reinforcement learning. Firstly, we built an AI supervisor with deep neural networks and generated simulated images based on configurations of the observation instruments and various observation conditions as prior information. Then, we trained the AI supervisor with simulated and real observation images, with the ground-truth catalogue and magnitudes of reference stars as the desired output. Upon completion of training, the AI supervisor can obtain the optimal parameters of the source-extraction algorithms for newly acquired images through automatically fine-tuning based on prior information about the observation conditions and on the properties of the observed star fields. Results . We evaluated the AI supervisor using simulated and real observation images. The results indicate that the AI supervisor effectively identifies the optimal parameters for the source-extraction algorithm in processing newly observed images within a few iterations. With these optimised parameters, the source-extraction algorithm achieves a higher photometry accuracy, higher precision rates, and a lower detection threshold. These enhancements underline the potential of the AI supervisor in fine-tuning source-extraction algorithms and other related astronomical data-processing algorithms.