A Workpiece Classification Model with ROI Adaptive Contour-driven Cropping
TAN Hanzhong
WEN Shuangbing
ZHANG Yedong
HUANG Haifeng
HU Tao
Abstract：To address the issue of redundant computation in non-region of interest(ROI)areas when existing workpiece classification models processed high-resolution images,a workpiece classification network model with ROI adaptive contour-driven cropping(ACDC-ClassNet)was proposed.This model utilized contour detection to localize the largest contour and its center within the image.Based on this,a standardized square ROI cropping region was generated,effectively eliminating background interference.A pre-trained residual network-50 layers(ResNet-50)model was adopted,and its classification head was adjusted to match the number of workpiece categories,enabling efficient feature focusing and classification.The results demonstrated that this ROI cropping strategy achieved an average reduction of 72.15%in the area of redundant regions,allowing the model to focus more effectively on workpiece details.Compared to the original ResNet-50,ACDC-ClassNet was observed to achieve improvements of 3.83,4.04,3.64,and 4.13 percentage points in accuracy,precision,recall,and F1-score,respectively.Furthermore,the strategy was also shown to outperform efficient network(EfficientNet)and vision transformer(ViT)models,achieving accuracy gains of 8.40 and 4.27 percentage points.ACDC-ClassNet provided a novel technical pathway for efficient visual inspection in industrial settings.
Keywords：workpiece classificationROI adaptive contour-driven croppingResNet-50transfer learningdetail perceptionintelligent manufacturingredundant features
Publication Date：2025-12-30
Online Publishing Date：2025-12-10(First online date of this platform, not the publication date of the document)
Pages：6( 517-522 )