Speaker
Description
With its extremely large field of view of ≃47 square degrees, which scans the entire northern sky every two nights, the Zwicky Transient Facility (ZTF) is a powerful tool for serendipitous detections of near-Earth asteroids (NEAs). This effort aims to both discover new NEAs and refine the orbital information of known objects, anticipating and mitigating future dangerous collisions. We present a novel pipeline for ZTF images that uses a convolutional neural network (CNN) to improve the detection capability of NEAs. Our work aims to minimize the dependency on human intervention of the current approach adopted by the ZTF. The target NEAs have high proper motions of up to tens of degrees per day and thus appear as streaks of light in the images. We trained our CNNs to find these streaks by using three datasets: a set with real asteroid streaks, a set with synthetic streaks and a set with a mix of real and synthetic, and tested the resultant models on a set of 115 real asteroids. The results achieved were almost identical across the three models: 0.843±0.005 for the completeness and 0.820±0.025 for the precision. No significant differences were found when assessing the quality of the detections; all models performed equally well when characterizing the streaks found in terms of position, angle with respect to the x-axis and length. The average error reported by the three pipelines was 1.84±0.03 pixels for the streak position, (0.817±0.026)° for the streak angle and -0.048±0.003 for the relative error in streak length. In addition, we compared the performance of our pipeline trained with a mix of synthetic and real streaks to that of the human scanners who vet ZTF candidate streak detections by analyzing a larger set of images containing 317 streaks flagged as valid by the scanners. Our pipeline achieved a precision of 99 %, detected 80 % of the streaks found by the scanners and 697 additional streaks that were verified to be real objects. In this case, the pipeline trained with a mix of real and synthetic streaks outperformed the pipeline trained only with real streaks by 10 % in completeness. This disparity in performance was explored to suggest possible reasons. Our results indicate that the new automated pipeline can complement the work of the human scanners at no cost for the precision and find more objects than the current approach. They also prove that the synthetic streaks simulated were realistic and can be used to enlarge training sets with insufficient real streaks or explore the simulation of streaks with unusual characteristics that have not yet been detected. Our pipeline not only shows a strong potential to make new findings in the ZTF data but also can be scaled up to other wide-field telescopes, setting the stage for fast and automated NEA discoveries in the next generation of astronomical surveys.