As maritime traffic continues to grow in complexity, testing ship collision avoidance systems has become increasingly challenging. Conventional methods for generating test scenarios often lack sufficient coverage of edge cases, limiting the effectiveness of collision avoidance algorithms in handling complex and critical situations. This paper introduces an enhanced generation method for edge test scenarios based on an improved seed-filling algorithm. By analyzing Automatic Identification System (AIS) historical data, typical ship encounter scenarios are extracted as seed scenarios. These seeds are expanded and diversified through a combination of random perturbations and multi-path expansion strategies. A reality and risk-based screening mechanism is applied to generate highly representative and high-risk edge test scenarios. Experimental results demonstrate that the proposed algorithm produces a broader range of diverse and hazardous scenarios, significantly improving the robustness and adaptability of ship collision avoidance system testing.