Machine learning-based automatic focusing for high magnification systems

Abstract

Precise focus plays a critical role in the quality of astronomical observations, significantly affecting scientific research. Accordingly, the automatic focusing of a high magnification telescope system is desirable. The auto-focusing level is measured using a focus measure operator. The principal challenge is to find a robust operator that can accurately estimate the focus level for high magnification astronomical images that suffer from high blur. In addition, images can suffer from aberrations as the distance from the image center increases, along with dramatic effects associated with dynamic observation conditions such as temperature, clouds, humidity, and seeing. In this study, we first propose a focus measure relying on fuzzy logic that can handle imprecise data. Second, we optimize the parameters of the fuzzy membership functions using multivariate particle swarm optimization (PSO). We apply the proposed method to five sequences of different star clusters acquired using the 74-inch telescope at the Kottamia astronomical observatory (KAO) during good seeing conditions. We then compare our proposed method with other traditional focus operators and rank them according to evaluation criteria. The results show that the proposed operator generally outperforms the others, and that the optimized operator provides further improvement.