Research on Automatic Recognition of Homologous Plastic Seals

Abstract

To realize the automatic identification of homologous plastic stamp, four classic convolutional neural network models were run by using the pytorch framework and three plastic stamps with the same chapter content were engraved using three laser scanning speeds to seal with uniform moderate pressure. 15 300 complete imprints were scanned to obtain a printed image as sample data. The effects of training sample size and network model on the automatic recognition of homologous seals were studied. The results show that the convolutional neural network can realize the automatic identification of homologous plastic stamp and the increase of the training sample size can increase the performance of the model. The highest test accuracy can reach 100%, in enough training sample conditions, Resnet50 model is the best choice.