Abstract: In order to improve the restraint that caused by the difficulty of obtaining abnormal data and the imbalance of data among various abnormal categories for current common anomaly detection algorithms, this paper proposes a semantic segmentation-assisted single teacher and multi student unsupervised anomaly detection algorithm. Multiple student networks were used in the knowledge distillation model to participate in prediction together, and generated the feature maps through statistical information such as the mean and variance of multiple student networks to enhance the stability and accuracy of the prediction results. By using the idea of feature pyramids, the method extracted multiple scales of visual features to improve the model's ability to detect abnormal regions of different scales. By using the boundary extraction capabilities of the semantic segmentation model, the semantic segmentation network was introduced to assist in correcting the anomaly detection results. Compared with the existing unsupervised anomaly detection algorithms, the proposed method has better ability in locating and detecting object surface anomalies in industrial scenes.