The application of image data in various fields is becoming increasingly widespread. Especially in security, the need for image detection is becoming more and more urgent. Anomaly detection, as an important aspect in Industrial Control Safety (ICS), aims to automatically identify and locate regions or objects in an image that deviate from normal pixels. This capability is crucial for enhancing the efficiency and precision of product safety testing systems. In response to this need, we propose a novel multi-scale endogenous enhanced anomaly detection model based on Generative Adversarial Networks (GANs). Our model operates on an unsupervised learning paradigm, allowing effective training even in scenarios with limited labeled data. By optimizing image details, our approach achieves robust anomaly detection. Furthermore, we introduce a multi-scale endogenous enhancement technique to bolster the model’s resilience. To quantify anomaly severity, we devise a unique anomaly scoring mechanism, providing decision support for the model’s assessments.We conduct experiments using the challenging MVTec dataset, demonstrating that our proposed method surpasses existing state-of-the-art approaches in both anomaly detection and localization.