Optimized trustworthy visual intelligence model for industrial marble surface anomaly detection

Abstract

In industrial manufacturing, to ensure the trustworthiness of visual intelligence, it is necessary that models can be able to detect surface anomalies and maintain product quality. In this context, this work provides a trustworthy visual intelligence model for industrial marble surface anomaly detection using EfficientNetV2B1 for feature extraction and LightGBM for classification. Using the Swarm Robotics Search and Rescue (SRSR) method, the hyperparameters of the model are tuned and provide an accuracy of 72.4% and an AUC of 0.89 on a marble surface dataset. Comparative study against models such as Gradient Boosting and XGBoost indicates the better performance of our technique in terms of accuracy, recall, and F1-score. These findings show its dependability for industrial quality control and provide an automated, dependable way to find marble surface flaws.