Safety Monitoring by a Graph-Regularized Semi-Supervised Nonnegative Matrix Factorization With Applications to a Vision-Based Marking Process

• Main Authors: Song Fan, Qilong Jia, Wan Sheng Cheng
• Format: Article
• Language: English
• Published: IEEE 2020-01-01
• Series: IEEE Access
• Subjects: Fault detection; nonnegative matrix factorization; semi-supervised learning; marking process
• Online Access: https://ieeexplore.ieee.org/document/9118902/

This paper proposes a brand-new method to perform safety monitoring using images for steel coil marking industrial processes. The new safety monitoring method is developed with the aid of a new graph-regularized semi-supervised nonnegative matrix factorization (GSNMF) algorithm. Compared with the existing nonnegative matrix factorization (NMF)-like algorithms, GSNMF is developed in an all-new manner so that it not only can take advantage of images with known labels and images with unknown labels to train a model for monitoring purpose, but also can take advantage of graph theory to improve the monitoring performance. Because any two different samples are connected by an edge in a graph, thus graph theory is beneficial for GSNMF to measure the similarity between any two different samples and to assign the same labels for the samples with close connections between them. As a result, GSNMF is more capable of analyzing the samples with a complicated distribution than the existing NMF-like algorithms, theoretically. Finally, an experiment on a steel coil marking process is adopted to evaluate the superiorities of our proposed method over the existing methods.