SLCNN: A Shipping Label Generation Model based on Convolutional Neural Networks

Shipping labels contain critical information that carriers need to deliver packages to their intended destinations. However, these labels are often damaged during packaging and delivery, resulting in unreadability, delivery delays, and significant losses. Existing techniques have been developed for improving shipping label detection and recognition; however, less attention has been given to generating shipping labels for evaluating sorting machines. In this paper, we propose SLCNN (a shipping label generation model based on convolutional neural networks) to generate shipping labels. SLCNN first captures the shipping labels from package images and classifies the unreadable ones into four types of groups. Next, the features of the unreadable ones are transferred to the readable ones to produce new shipping labels. In the experiment, SLCNN and two other models (i.e., DCGAN and DIP) each generated datasets of 9,000 shipping labels. The results indicate that the label dataset generated by SLCNN exhibited greater distinguishability, leading to shorter recognition times for the sorting machine. Additionally, the shipping label dataset from SLCNN contains more textual features and richer informational content.