“In recent years, sudden surface anomalies have occurred frequently. To address the problem of insufficient annotated samples in the early stages of sudden anomalies, a lightweight small sample surface anomaly detection method based on a visual language pre training model is proposed, providing a solution to the problem of remote sensing emergency monitoring.”