Anomaly Detection of Industrial Images Based on Instance Segmentation with Support Vector Machines Model

In smart manufacturing systems, anomaly localization plays a crucial role in enhancing intricate production processes. However, When it comes to identifying normal zones around anomaly identification in high-dimensional points and industrial images, the models perform poorly. In this study, in order to address the aforementioned problems To find the abnormality in an industrial image, Instance Segmentation with Support Vector Machine (IS-SVM) are used. Initially, padding techniques are used to resize the input pictures of the MvTec AD dataset in order to improve segmentation. The instance segmentation approach is applied to accurately identify if a picture flaw is abnormal or not. In the end, the SVM classifier produces a prediction of the industrial image anomalies that is more accurate by Using the Histogram Of Orientation features. The developed model then gets over the restrictions by improving generalization ability and decreasing the false detection rate by using the IS-SVM architecture with a 98.97% mean ROCAUC, the constructed model outperforms the existing models like Grey Correlation-based Picture Fuzzy-Evaluation (GCPF), Patch DistrIbution Modeling (PaDiM-R18-Rd100), Memory-based Segmentation Network (MemSeg), Reliable Anomaly Detection and Localization System (RADL), and Vision Transformer Network Anomaly Detection and Localization (VT-ADL).