The work aims to propose an improved reagent card printing defect detection algorithm YOLOv5s-EF based on deep neural network YOLOv5s to solve the problems of low efficiency, high cost and easy to miss detection in manual sorting of reagent cards with printing defects in reagent card manufacturers. High-quality defect image data sets were obtained by image preprocessing algorithm. Efficient Channel Attention (ECA) mechanism was added to the backbone feature extraction network of YOLOv5s to enhance the representation ability of important features in feature maps. Focal loss function was introduced to alleviate the influence of imbalance between positive and negative samples. Combined with the positioning results of the printing area, a method of similarity matching of feature vectors was proposed, which was based on the quadratic accurate positioning and the construction of azimuth feature vectors. The experimental results showed that the average detection accuracy of the reagent card printing defect detection algorithm proposed in this paper could reach 97.3% and the speed was 22.6 FPS on the test set. Compared with other network models, it can identify and locate various printing defects. The model has good detection speed and robustness, which is beneficial to improve the intelligent level of enterprise production.