Research on X-ray Image Fusion Algorithm for Food Foreign Object Detection

Abstract

In the field of non-destructive testing of foreign objects in food, the high cost and low efficiency of manual labeling greatly limit the application of X-ray foreign object detection systems. To overcome this problem, this paper proposes a technique for fusing foreign object images with food images, and the fused images enable the automatic labeling of foreign objects. Firstly, X-ray images of foreign objects and food images were collected, and data augmentation was performed on the foreign object images to increase their diversity. Then the food images were fused with the enhanced foreign object images, and the foreign objects were automatically labeled in the fused food images. Finally, the foreign object detection models Model_Y2 and Model_Y1 were established using the dataset automatically annotated by the image fusion method and the dataset manually collected and annotated by traditional methods. The results demonstrate that the proposed method substantially decreases annotation time by 90% while concurrently improving annotation efficiency and accuracy. Comparatively, Model_Y2 outperforms Model_Y1 with a 4.5% higher mAP@0.5:0.95. This indicates that the method not only enhances data annotation efficiency and quality but also improves the accuracy of X-ray foreign object detection, providing a highly efficient and practical technical solution for the intelligent development of food safety inspection.