DrtNet: An Improved RetinaNet for Detecting Beverages in Unmanned Vending Machines

摘要

With the rapid development of technologies such as mobile payment, deep learning, and cloud computing, the transformation of traditional retail models has become possible. Research on deep learning algorithms has become a hot direction in computer vision. Deep learning techniques have been widely used in face recognition, intelligent transport system, smart city, etc. At present, smart unmanned vending machines (UVMs) are of one of the representative carriers of unmanned retail business models. Compared with traditional vending machines, they have greatly improved user experience and saved implementation costs. One core issue of smart UVMs is how to quickly and accurately identify which and what kind of items a user is taking. In this paper, deep learning is used to solve the problem of object recognition in smart UVMs. To accurately obtain the information of products that a user may take from a UVM, an object detection-based method is used. Specifically, an object detection model, namely DrtNet, is proposed and designed under the smart UVM scenario. The backbone network of DrtNet adopts deformable convolution and group normalization layers. The used loss function contains focal loss function and balance Ll loss function, which can largely improve the recall rate of beverage detection. Experimental results demonstrate the feasibility and effectiveness of applying the deep learning model to smart UVMs.