Algorithm for cigarette carton label target recognition based on YOLOv3

To address the issues affecting the detection effectiveness of carton label images in cigarette carton logistics, such as dense and overlapping targets and significant disparities in shape and aspect ratio, this paper proposes an enhanced QR code target detection algorithm based on YOLOv3. By integrating deformable convolution for feature extraction, adopting the Soft-nms algorithm for more flexible non-maximum suppression, and incorporating the focal loss function, the modified YOLOv3 algorithm demonstrates improved performance. The algorithm's efficacy is validated through comparative experiments on a specific dataset for cigarette carton labels, showing a 3.37% increase in F1-score sample model identification rate and a 2.2% improvement in target detection accuracy. These enhancements offer a targeted and effective solution for increasing the accuracy of cigarette carton label recognition.