- Conference Article
- 10.1109/eiecs67708.2025.11283231
Piston Surface Defect Detection Based on Improved Faster R-CNN
- Sep 26, 2025
- Jiarong Yang + 1 more +1
To improve the detection rate and reduce the false detection rate of tiny defects on piston surfaces, a small target detection algorithm based on the improved Faster R-CNN is proposed. First, ROI region cropping is performed on original piston images to increase the proportion of small targets. Subsequently, the cropped images are input into the improved Faster R-CNN, which incorporates the FPN (Feature Pyramid Network), Guide Anchor RPN (Guided Anchoring Region), and Double-RoI Head dual-head detection structure. Under the PyTorch framework, comparative experiments are conducted among the improved Faster R-CNN, YOLOv8, YOLOv12, YOLONAS, RT-DETR, and the original Faster R-CNN. The results show that the recall@0.3 and precision@0.3 of the improved Faster RCNN are $81.1 \%$ and $64.6 \%$ respectively, and its small target detection capability is significantly superior to other models. Combined with the edge contour detection algorithm designed in this study, it can further detect extremely tiny edge defects missed during the convolution process, and meeting the detection requirements of industrial scenarios.
Read more