Wind power curve is a core characteristic describing the relationship between wind speed and turbine output power, which is crucial for wind farm operation monitoring and anomaly detection. However, traditional single-feature analysis methods based on Supervisory Control and Data Acquisition (SCADA) systems cannot fully characterize the abnormal features of power curves. Meanwhile, data scarcity and environmental variations restrict detection performance. To solve these problems, a wind power curve anomaly detection method fusing image-topology-semantic multi-feature fusion and contrastive learning is proposed. This method comprehensively captures abnormal patterns of power curves from multiple perspectives by integrating image, topological and semantic feature representations. A contrastive learning mechanism is introduced to improve feature discriminability in low-label scenarios. The effectiveness of the proposed method is verified using SCADA operation data from multiple wind turbines in actual wind farms. Comparative experimental results show that the detection accuracy of the proposed method is significantly superior to existing methods.