Casting defect detection based on improved DETR

Abstract

Casting products are widely used in many industrial fields. Castings are prone to defects in the production process. Due to the high utilization rate of castings, the quality of castings has become the focus of industrial production. Therefore, casting defect detection is very important in the production process. This paper proposes a casting defect detection method based on improved DETR to improve detection accuracy. Firstly, multi-scale feature fusion is added in the feature extraction stage; then, the ECA-NET attention mechanism module is introduced to improve the backbone network; secondly, the attention mechanism in the Transformer module is improved by using relative position encoding; finally, in the casting data, the improved algorithm is trained and tested on the set. The experimental results show that the accuracy of the improved DETR model proposed in this paper can reach 49.6%, which is 6.3% higher than the original DETR algorithm. Compared with other mainstream target detection algorithms, the casting defect detection model proposed in this paper can effectively identify Casting defects and achieve high detection accuracy.