Meta Learning-Driven Iterative Refinement for Robust Anomaly Detection in Industrial Inspection

This study investigates the performance of robust anomaly detection models inindustrial inspection, focusing particularly on their ability to handle noisydata. We propose to leverage the adaptation ability of meta learning approachesto identify and reject noisy training data to improve the learning process. Inour model, we employ Model Agnostic Meta Learning (MAML) and an iterativerefinement process through an Inter-Quartile Range rejection scheme to enhancetheir adaptability and robustness. This approach significantly improves themodels capability to distinguish between normal and defective conditions. Ourresults of experiments conducted on well known MVTec and KSDD2 datasetsdemonstrate that the proposed method not only excels in environments withsubstantial noise but can also contribute in case of a clear training set,isolating those samples that are relatively out of distribution, thus offeringsignificant improvements over traditional models.