Advances in deep learning-based image recognition of product packaging

Abstract

The information about the goods on the shelves needed to be obtained by the dealers in real time, so that they could serve the customers better. The visual-based shelf commodity detection methods have attracted extensive attention in recent years. Among them, the target detection method based on deep learning could automatically understand the picture features, which greatly promoted the development of goods shelf identification. This paper made a comparison and analysis between the common target detecting data set and the data set of goods on shelf from their features, advantages and application. It came up with a method to build an excellent data set of goods, and sorted out the data enhancement methods to perfect the data set for defective data sets. Then, from the three angles of large-scale package identification, small target identification and partial occluded recognition, the latest product package identification method was summarized. At last, the limitations of deep learning method in product identification were discussed and the future development direction was looked forward.