Surface Defect Detection Algorithm of Aluminum Based on Improved Faster RCNN

Abstract

There are some problems in the surface defect detection of industrial aluminum products, such as small defect samples, extreme length-to-width ratio of defect, low precision of small defect detection, etc. To solve these problems, an aluminum surface defect detection algorithm is proposed based on improved Faster RCNN. The number of defect samples is increased by data augmentation, and the residual network ResNet50 is employed as the backbone feature extraction network to extract aluminum defect features. Then the path enhancement feature pyramid network (PAFPN) is added to the backbone feature extraction network to form a multi-scale feature map which strengthens the utilization of feature information from the lower layers. Soft non-maximum suppression (Soft-NMS) is used to further improve the detection performance of the algorithm. Results show that the mean average accuracy (mAP) of the proposed algorithm is 78.8%, which is 2.2% higher than the original algorithm.