Research on Product Detection Algorithm for Intelligent Refrigerator

In this paper, we designed a pure visual static identification scheme for the intelligent freezer system. The camera is added to each layer in the freezer to detect and identify the product images before and after the user switches the freezer door. The difference in the quantity of the product is calculated as the amount of consumption. In order to solve the problem of real-time detection of many types of products, this project designed a lightweight network structure IceboxNet suitable for product detection and recognition. In the visual intelligent freezer scene, this structure can not only achieve high accuracy recognition of hundreds of commodities. Moreover, the detector based on the SSD algorithm can greatly reduce the depth model size and improve the algorithm speed without loss of precision. In the improved scheme of SSD algorithm, considering the computing power of the device, IceboxNet is used as the basic network to simplify the multi-layer feature fusion mechanism. In order to further improve the accuracy of product detection, this paper sets the size and length of the default proposal for the product dataset. The width ratio and the model parameters in the hundred kinds of product identification are used to initialize the parameters of the object detector. Finally, the model designed in this paper has a mAP of 99.15% in the product data set and a detection speed of 74FPS.