Industrial anomaly detection with multiscale autoencoder and deep feature extractor-based neural network

With the maturity of deep learning image recognition technology and the popularity of automated production lines, deep learning industrial anomaly detection has become an important research topic in recent years. In this study, an anomaly detection model with a multi-scale autoencoder and deep feature extractor is proposed. This model was confirmed to obtain the highest area under the curve (AUC) in 14 of the 17 industrial detection tasks. In addition, the receiver operating characteristic (ROC) curves show that an appropriate threshold of the proposed model exists, which can achieve a low false-positive rate and maintain a high true-positive rate. Furthermore, the influence of different feature extractors on the method is discussed. It was shown that the proposed method can maintain good detection ability with most of the feature extractor. Therefore, it is suitable for industrial optical inspection systems with different hardware conditions.