Automated Steel Surface Defect Detection using Optimized Cascaded CaffeNet Region Network
1EEE, Dr. M.G.R Educational and Research Institute, India
2EEE, K.S. Rangasamy College of Technology, India
3CSE, SRM Institute of Science and Technology, India
4ECE, Saveetha Institute of Medical and Technical Sciences, Chennai, India., India
*Author to whom correspondence should be addressed:
E-mail: sujatha.kesavan.example@university.edu (SK)
E-mail: sujatha.kesavan.example@university.edu (SK)
Received: November 23, 2024 | Revised: April 26, 2025 | Accepted: May 20, 2025 | Published: June 2025
Abstract
Only a small number of low-resolution images are available to address the fundamental difficulties in defect detection during steel sheet manufacture, which results in less than ideal recognition accuracy and overall deep learning algorithm performance. A deep learning-based approach for detecting steel surface flaws is described in this paper. A new Optimized Cascaded CaffeNet Region Network (OCAFRN) is utilized to detect defects from surveillance cameras by utilizing vision machine technology. By mapping the features using R-CNN and the Electric Fish Optimization Approach (EFOA), this cascaded structure with a 3D CaffeNet is utilized to extract the region of the steel sheets with faults. It outperforms the other standard algorithms in terms of recognition accuracy, which is consistent and comprehensive for both positive and negative categories under a variety of scenarios. In contrast, OCAFRN, which is an improved network model, exhibits optimal performance for smaller image models, achieving 96.5% of defect detection accuracy with a reduction in 3.2% of Mean Absolute Error (MAE) in comparison with the existing methods like Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Back Propagation Algorithm (BPA) and Radial Basis Function Network (RBF).
Keywords
Accuracy ; Defect Detection ; CaffeNet ; Electric Fish Optimization ; Steel Sheets
Available Repositories
Share Article
Article Metrics
--
Views
--
Downloads
--
Citations
Export Citation
Full Text
