Atrous Convoluted VGG based Metal Surface Defect Classification

Classifying metal surface flaws is essential to ensuring the safety and quality of commercial products. Conventional techniques frequently depend on time-consuming visual inspections, which can be ineffective and subject to human error. This research presents a deep learning model based on the Atrou...

Full description

Saved in:
Bibliographic Details
Published in2025 International Conference on Advances in Modern Age Technologies for Health and Engineering Science (AMATHE) pp. 1 - 6
Main Authors Preethaa, K.R. Sri, Devi, M. Shyamala, Natarajan, Yuvaraj, Shanmugapriya, K.
Format Conference Proceeding
LanguageEnglish
Published IEEE 24.04.2025
Subjects
Accuracy
atrous
augmentation
CNN
Computational modeling
convolution
Feature extraction
filtering
Information filters
Inspection
Metals
Safety
Surface texture
Surface treatment
VGG
Visualization
Online AccessGet full text

Cover

More Information
Summary:Classifying metal surface flaws is essential to ensuring the safety and quality of commercial products. Conventional techniques frequently depend on time-consuming visual inspections, which can be ineffective and subject to human error. This research presents a deep learning model based on the Atrous Convoluted VGG (AC-VGG16) to improve the classification of metal surface defects. The Metal surface defect Dataset that includes 1200 metal surface images was used for implementation. The AC-VGG16 model initiates by organizing the metal surface images for formation of labelled images. Then labeled images are subjected to create \mathbf{2 5, 2 0 0} augmented images. The augmented images are again processed to form filtered images which are applied to the existing CNN and proposed ACVGG16 model to analyze the performance. The proposed approach enhances the network's capacity to capture multi-scale information by integrating atrous convolutions into the VGG architecture, which is especially advantageous for identifying irregular flaws in metal surfaces. With its dilated convolutions, the improved VGG model maintains computational efficiency while allowing for larger receptive fields, improving feature extraction from intricate patterns and textures frequently observed on metal surfaces. The implementation shows that ACVGG16 model with dilated vessel filtered metal surface images outperforms with the high accuracy of \mathbf{9 9. 4 3 \%} towards metal surface defects classification.
DOI:10.1109/AMATHE65477.2025.11080947