A Few-Shot Complex Chip Defect Detection Algorithm Guided by Priori Boxes

chip surfaces are characterized by complex textures, and defect samples are limited in variety, making it difficult to ensure the accuracy of defect detection. to address this issue, this paper proposes a defect detection algorithm guided by prior boxes. first, different weights are assigned based on the detection priority of different chip regions and the consistency of positive samples, guiding the feature extraction network to selectively extract positive sample features. then, a small number of real defect samples are used to fine-tune the object segmentation model, and a fusion detection model is constructed based on the prediction probability box of the object segmentation network and the similarity between the test sample and the positive sample features. finally, experiments are conducted on a real dataset collected from a chip production line. the experimental results demonstrate that, in terms of few-shot complex chip surface defect detection, this method improves accuracy compared to traditional defect detection algorithms.