An Improved Correlation Filtering Method for Tracking Maritime Small Targets of GF-4 Staring Satellite Sequence Images

Continuous, accurate and real-time earth observation plays an increasingly important role in battlefield situation awareness, environmental monitoring and many other fields. In this paper, GF-4 staring satellite’s sequence images are taken as the research objects, and its features such as “short imaging interval, long image sequence and high resolution” are used to solve the tracking problem of large and medium ships at sea. Using the target information, the dark-channel background modeling method and multi-scale Retinex illumination algorithm are improved to achieve target enhancement. The multi-scale features of the target based on time and space information are extracted, and the target tracking is realized by improved kernel correlation filtering algorithm. By comprehensively utilizing unscented Kalman filter model and target significance features, tracking results are optimized to improve tracking accuracy. The experimental results show that the proposed algorithm can track dim and small targets in the staring satellite sequence images effectively, and it has reference significance for the application of similar high-precision satellites.