Abstract:This study aims to improve the accuracy and efficiency of detecting defects on polished metal surfaces by developing a new detection system. A network model explicitly designed for detecting defects on polished metal surfaces and a convolutional feature concentration and activation network (CFCANet) are proposed in this paper. This model significantly enhances the recognition of tiny defects by introducing a small-target detection head, ensuring high-precision detection results. In addition, the feature concentration and activation (FCA-C2f) module proposed in this study enhances the model's sensitivity to anisotropic features, thereby improving defect detection accuracy. The content-aware reassembly of features (CARAFE) upsampling algorithm is used instead of traditional nearest-neighbour interpolation methods to effectively preserve detailed information and improve the quality and efficiency of upsampling. By optimising the lighting conditions and using composite light source illumination technology, the probabilities of missed detections and false alarms can be reduced. Combined with the CFCANet detection network, the defect detection performance of the proposed method on polished metal surfaces is effectively enhanced. To validate the effectiveness of the proposed method, a new dataset for detecting defects on polished metal surfaces, PMS-DET, was constructed in this study and validated on the NEU-DET dataset. Experimental results show that CFCANet effectively improves the defect detection accuracy on polished metal surfaces, achieving a mAP0.5:0.95 value of 42.4 % on the PMS-DET dataset, an increase of 11.9 %. The model parameters are reduced by 6.7 %, and the detection speed is improved by 28.1 %. Compared with existing detection models, this research method demonstrates significant improvements in detection accuracy, model size and computational efficiency, especially regarding GFLOPs and detection speed, proving its potential application value in practical industrial scenarios.

An Improved Feature Enhancement CenterNet Model for Small Object Defect Detection on Metal Surfaces

MDD-Net: A novel defect detection model of material microscope image

Metal surface defect detection based on grouped convolution and bidirectional feature fusion

Real-time Detection System for Polishing Metal Surface Defects Based on Convolutional Feature Concentration and Activation Network

A Sub-Region Unet for Weak Defects Segmentation with Global Information and Mask-Aware Loss.

CenterNet-based defect detection for additive manufacturing

Metal Surface Defect Detection Method Based on Improved Cascade R-CNN

Defect Detection in Steel Using a Hybrid Attention Network

MeDERT: A Metal Surface Defect Detection Model.

A deep learning model for steel surface defect detection

ADDet: An Efficient Multiscale Perceptual Enhancement Network for Aluminum Defect Detection

LE–MSFE–DDNet: a defect detection network based on low-light enhancement and multi-scale feature extraction

STFE-Net: A multi-stage approach to enhance statistical texture feature for defect detection on metal surfaces

An efficient multiscale enhancement network with attention mechanism for aluminium defect detection

SCFNet: Lightweight Steel Defect Detection Network Based on Spatial Channel Reorganization and Weighted Jump Fusion

A Study of Defect Detection Techniques for Metallographic Images

DANet: Enhancing Small Object Detection through an Efficient Deformable Attention Network

Attention Pyramid Dilated Region-based Model for Metallurgical Defect Detection

Unleashing the power of AI in detecting metal surface defects: an optimized YOLOv7-tiny model approach

LAACNet: Lightweight adaptive activation convolution network-based defect detection on polished metal surfaces

An Enhanced YOLOv4 Model with Self-Dependent Attentive Fusion and Component Randomized Mosaic Augmentation for Metal Surface Defect Detection