A Multi-Stage Swarm Intelligence based Optimization for Deep Feature Reduction in Metal Surface Defect Classification

Abstract Automated metal surface defect detection is critical for industrial quality assurance, where early and accurate identification of imperfections ensures product reliability, reduces operational losses, and supports sustainable manufacturing practices. In this study, we propose a novel multi-stage swarm intelligence optimization framework for deep feature reduction in metal surfaces. Here, we aligned with the Industry 5.0 principles by integrating intelligent automation with bio-inspired and human-centered, sustainable industrial processes. Initially, pretrained Convolutional Neural Networks (CNNs), including ResNet50 and EfficientNet-B0, are utilized as fixed feature extractors to convert raw surface images into high-dimensional numerical representations. Then, Principal Component Analysis (PCA) is employed to reduce feature dimensionality up to a constant components. Later, the swarm-based optimizing algorithms, which include Ant Colony Optimization (ACO) and Artificial Bee Colony (ABC), were applied to select the most discriminative subset of features dynamically. Before evaluation, these optimized features were then input into classical machine learning classifiers, including XGBoost (XGB) and Random Forest (RF), that were later evaluated across multiple metrics such as Accuracy, Precision, Recall, F1-Score, Matthews Correlation Coefficient, and Balanced Error Rate. This optimized pipeline utilized merely 185–244 discriminative features at the end of the pipeline and achieved an accuracy of 99.3%. Additionally, Grad-CAM visualizations provide explainable insights into the regions that are contributing most to predictions, which enhances interpretability for industrial stakeholders. Thus, this proposed approach demonstrates a robust, computationally efficient, and explainable solution for automated defect inspection, supporting sustainable, high-quality manufacturing processes which is in line with the United Nations Sustainable Development Goal 9 (SDG 9) and the vision of Industry 5.0.