Unsupervised Tablet Defect Detection Method Based on Diffusion Model

Reconstruction-based unsupervised detection methods have demonstrated strong generalization capabilities in the field of tablet anomaly detection, but there are still problems such as poor reconstruction effect and inaccurate positioning of abnormal areas. To address these problems, this paper proposes an unsupervised Diffusion-based Tablet Defect Detection (DTDD) method. This method uses an Assisted Reconstruction (AR) network to introduce original image information to assist in the reconstruction of abnormal areas, thereby improving the reconstruction effect of the diffusion model. It also uses a Scale Fusion (SF) network and an improved anomaly measurement method to improve the accuracy of abnormal area positioning. Finally, the effectiveness of the algorithm is verified on the tablet dataset. The experimental results show that the algorithm in this paper is superior to the algorithms in the same field, effectively improving the detection accuracy and abnormal positioning accuracy, and performing well in the tablet defect detection task.