Food packaging defect detection by improved network model of Faster R-CNN

Abstract

Objective: Accurate identification and location of paper packaging box defects. Methods: The improved network model of Faster R-CNN was applied to automatically detect box defects. The data of the training set picture was enhanced and noise was added to improve the training accuracy and robustness of the model. The feature extraction network was replaced with ResNet50, and the feature pyramid network （FPN） was fused to improve the multi-scale detection ability of the model. K-means++ was used to cluster the defect scale in the dataset and optimize the anchor box scheme. Results: The average accuracy （AP） of the improved Faster R-CNN model on the test set reached 93.9%, and the detection speed reached 8.65 f/s. Conclusion: The improved Faster R-CNN model can effectively detect and locate box defects, which can be applied to the automatic detection and sorting of box defects.

Keywords

• faster r-cnn
• resnet50
• defect detection
• feature pyramid network (fpn)