To solve the problems of picking errors of strip tobacco by sorting machine and manual sorting in tobacco logistics center, the work aims to propose a strip tobacco identification model with faster convergence speed and higher accuracy based on YOLOv5s algorithm from the perspective of ensuring the real-time performance and recognition accuracy. Firstly, CA attention module was integrated into the YOLOv5s network architecture to better extract features and improve the accuracy of target acquisition. Then, the nearest neighbor in the original network was changed to lightweight general upper sampling operator CARAFE to obtain a larger feeling field. Next, the Ghost module was embedded in the backbone network to lightweight the network. Finally, the tobacco image acquisition system was built in the tobacco logistics center to establish the tobacco image data set. Compared with YOLOv5s, the proposed optimization algorithm was reduced by 45.8%, mAP@0.5 reached 99.3%, and the recognition rate was about 99.9% on the strip tobacco error correction system. The optimization algorithm proposed can meet the requirements of high-speed strip tobacco sorting and identification with high accuracy.