Product Recognition for Unmanned Vending Machines.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Unmanned retail, particularly using intelligent unmanned vending machines (UVMs), is a rapidly growing market.
• Current product recognition methods struggle with large-scale categories and accuracy in UVMs.

Purpose of the Study:

• To develop an effective method for large-scale product recognition tailored for intelligent UVMs.
• To address the limitations of existing recognition systems in terms of category scalability and accuracy.

Main Methods:

• Utilized manifold learning to analyze product similarities and differences.
• Developed a hierarchical multigranularity label to guide representation learning.
• Proposed a hierarchical label object detection network with a coarse-to-fine refine module (C2FRM) and multiple granularity hierarchical loss (MGHL).
• Collected and utilized the GOODS-85 dataset, specifically designed for UVM scenarios.

Main Results:

• The proposed method effectively mines product similarities within large categories.
• Hierarchical multigranularity labels optimize the learning process.
• Experimental results demonstrate significant improvements in product recognition accuracy for UVMs.

Conclusions:

• The developed method enhances product recognition for large-scale categories in intelligent UVMs.
• The approach offers a promising solution for the growing unmanned retail market.
• The GOODS-85 dataset provides a valuable resource for future research in this domain.