The work aims to propose a method of applying digital twin technology in tobacco packaging to solve the problem that at the current tobacco industry, there is a lack of automation in the production of tobacco roasting drop inspection and industrial opening robots, so that it is difficult to determine whether there is a yellow cardboard sandwich material in the tobacco packaging box. Through the construction of a complex tobacco packaging sandwich material digital twin model, image processing was used in combination with the convolutional neural network to get the complex tobacco packaging sandwich material detection method based on machine vision. The system could calibrate the shape of the straw board sandwich and determine its center, and enable the industrial robot to accurately locate the sandwich material of the tobacco packaging box, automatically increase or decrease materials when packing, and achieve effective digital material information tracking. Experiments showed that the accuracy of the robot in determining the sandwich material of the box was as high as 94.88%, and the efficiency was better than that of manual checking. It can save human resources to a certain extent and improve the production efficiency of the factory.