Critical space assets require continuous monitoring to prevent potential losses. Auxiliary satellites protect these assets by observing and tracking approaching targets. Although observation satellites can make rapid autonomous onboard decisions, they face challenges due to limited computational capacity. The two mainstream command and control methods currently available do not meet the demands of onboard decision-making. Highly procedural decision-making methods require extended decision times, while rapid-response intuitive or heuristic methods carry significant error risks. To address this, this paper proposes a multi-granularity decision-making method for optical space surveillance satellites. First, multi-granularity relative orbit determination algorithm models and multi-granularity impulsive orbit manoeuver algorithm models were developed. Based on these models, a granularity selection method for sequential three-way decisions is proposed. In non-emergency situations, fine-granularity models are preferred to conserve fuel, while in emergency situations, coarse-granularity models are used to enhance decision-making speed and reduce positional deviations caused by the manoeuvering game. In random multi-scenario tests, the proposed method demonstrates lower average terminal positional deviations and fuel consumption compared to single-granularity (highly procedural or rapid-response) and random-granularity methods.