A Lightweight Recognition Method for Three-Segment Codes on Express Waybill Under Arbitrary Poses

Aiming at the low parcel sorting efficiency at terminal logistics sites, a lightweight recognition method for three-segment codes on express waybill is proposed. To quickly and accurately locate three-segment codes from captured images in arbitrary poses, an orientation-aware express waybill detection algorithm is designed by adding keypoint detection branch to the rotated object detection network YOLOv8n-OBB. The improved model can directly obtain upright express waybill images, from which a lightweight object detection network, YOLOv8n, is employed to extract sub-images of the three-segment code regions. To address the issue of recognition errors encountered by the lightweight OCR network SVTR-T for three-segment code recognition, an improved SVTR-T network is designed by embedding a CvT module to enhance its ability to distinguish the local features of similar characters and promote the recognition accuracy. Experimental results show that the proposed method can efficiently obtain the position, orientation and recognition result of express waybill, achieving a mAP50 of 0.995 for express waybill detection, an accuracy of 0.991 for orientation classification, and an accuracy of 0.845 for three-segment code recognition, and that the total time from detection to recognition is 54 ms, meeting the requirements for efficient parcel sorting.