To improve the application of Faster R-CNN model in metal surface scratch detection
HUO Chunbao
WANG Tongli
TONG Zhibo
Abstract：Aiming at the limitations of existing technologies in scratch detection on metal surfaces,an improved two-stage algorithm model based on Faster R-CNN is proposed.This model introduces a feature pyramid structure into the Faster R-CNN framework to enhance the ability of multi-scale feature extraction.Then,it optimizes the feature pyramid by incorporating a frequency-weighted noise mechanism,which further improves the feature representation capability.Finally,to address the irregularity of scratch shapes,a deformable convolution module is introduced,which significantly improves the model's detection performance for diverse scratch features.Experiments were conducted using the same dataset for training.By comparing the key indicators such as precision,accuracy,and recall rate of the model before and after improvement.The experimental results show that the improved Faster R-CNN has achieved significant improvements in detection efficiency and accuracy.The mean average precision(mAP)reaches 93.1%,which fully validates the effectiveness and practicality of the model.It provides the best solution for addressing the limitations of current metal surface scratch detection.
Keywords：deep learningdeformable convolutionfeature pyramidtwo-stage algorithm
Publication Date：2025-03-31
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：10( 71-80 )