Product type recognition from visuals on beverage bottle cap with deep learning

Beverages produced in industrial beverage production lines are filled in bottles, bottle caps are closed and delivered to consumers in order to protect their freshness and integrity. Beverage companies may prefer different visual designs or figures on bottle caps for each product type or Stock Keeping Unit (SKU) because of marketing and advertising strategies or differentiating one product regarding to another in their product range. Therefore, it is important to use the right cap design on the right product. Due to a problem to be experienced in the production process, it is possible to apply a cap that does not belong to the product that is currently being produced in the production line, in this case a quality problem that needs to be solved for the manufacturer arises. In this study, in order to solve mentioned problem, it is aimed to recognize the visuals on the beverage bottle cap by Deep Learning, accordingly a relationship is established between the cap and the product type. Convolutional neural network is a class of deep learning techniques. VGG-16 model, which is a convolutional neural network model, has been used to recognize the product type. Except for the last layer of the fully connected layer, transfer learning has been made over the model's training with the ImageNet dataset. Classification layer, which is the last layer of the fully connected layer, has been retrained with a data set consisting of a total of 7200 images of 9 different products created for this study. In the training and validation process, a satisfactory result has been achieved by achieving 99.99% training accuracy. In addition, a prototype software has been prepared to recognize product type of the cap belonging to different product type and prevent to apply on the bottle.