Interpretability-Aware Industrial Anomaly Detection Using Autoencoders

The past decade has witnessed wide applications of deep neural networks in anomaly detection. However, the dearth interpretability often hinders their reliability, especially for industrial where practical users heavily rely on interpretable methods to provide explanations decision-making. In this paper, we propose a reconstruction-based approach unsupervised detection anomalies defect data. Our algorithm employs an score during both training and test phases. Specifically, train autoencoder with loss function that incorporates interpretability-aware error term. After training, processes specific feature from difference between image average images produces attention map is used detecting anomalies. method not only achieves competitive performance compared non-interpretability-aware but also maps facilitate direct explanation results, which can potentially be useful practitioners.