Research on Surface Defect Detection of Microchannel Aluminum Flat Tubes Based on Improved Faster RCNN

To address the challenges of low detection rates for surface defects in microchannel aluminum flat tubes and poor detection effectiveness for small target defects using existing deep learning methods, a Faster RCNN based detection method is proposed. To overcome the limitations of multi-scale recognition in Faster RCNN, the defects were addressed by replacing the feature extraction network with ResNet152 and integrating FPN. These enhancements significantly improved the detection of intricate defects. The experimental results reveal that the improved model achieves an impressive mean average precision of 96.1% in detecting surface defects of microchannel aluminum flat tubes, which is 6.4% higher than the original Faster RCNN, and the detection accuracy of scuffing, scratches, and dirty spot defects is improved by 7.0, 8.9, and 4.1%, respectively. Moreover, the proposed model surpasses alternative target detection algorithms, maintains a low false detection rate, and significantly improves the detection of small target defects on microchannel aluminum flat tubes.