Mechanical Parts Detection Algorithm Based on Enhanced Faster R-CNN

Industrial target detection has been a popular field of target detection, because it can improve the efficiency of industrial production intuitively. Many scholars at home and abroad have achieved many research results till now, but there are still some limitations in the traditional algorithms. Many algorithms can only adapt to the industrial manufacturing conditions partially, and cannot be applied well when changing to another industrial environment. Deep learning has exploded in popularity in the field of target detection during the last several years, but its application to industrial target detection is not mature. We postulate an enhanced Faster R-CNN-based industrial target localization algorithm for mechanical parts in this research. We constructed our own mechanical parts dataset by shooting sampling, and improved Faster R-CNN target detection algorithm in five aspects. The target detection model was trained by migration learning. The experimental findings showed that our industrial target identification method based on enhanced Faster R-CNN outperforms the classic VGG16 Faster R-CNN model and Resnet50 Faster R-CNN model in terms of target detection accuracy.