Direct satellite-to-device communication is a promising future direction due to its lower latency and enhanced efficiency. However, intermittent and unpredictable terrestrial interference significantly affects system reliability and performance. Continuously employing sophisticated interference mitigation techniques is practically inefficient. Motivated by the periodic idle intervals characteristic of burst-mode satellite transmissions, this paper investigates online interference detection frameworks specifically tailored for satellite-to-device scenarios. We first rigorously formulate interference detection as a binary hypothesis testing problem, leveraging differences between Rayleigh (no interference) and Rice (interference present) distributions. Then, we propose a cumulative sum (CUSUM)-based online detector for scenarios with known interference directions, explicitly characterizing the trade-off between detection latency and false alarm rate, and establish its asymptotic optimality. For practical scenarios involving unknown interference direction, we further propose a generalized likelihood ratio (GLR)-based detection method, jointly estimating interference direction via the Root-MUSIC algorithm. Numerical results validate our theoretical findings and demonstrate that our proposed methods achieve high detection accuracy with remarkably low latency, highlighting their practical applicability in future satellite-to-device communication systems.