Software-Based Shielding: un approccio di machine learning per risolvere il problema di accuratezza inventariale nel retail RFID

Abstract

Radio-Frequency Identification (RFID) technology has been widely used primarily for tracking and identifying objects in supply chains. The fashion retail industry has been leading its adoption, primarily for the manifold use cases the technology enables for manufacturers, retailers and end consumers. These use cases range from increased productivity and accuracy in inbound outbound processes, to inventory accuracy and replenishment from the backroom in the stores. RFID deployments in fashion stores typically leverage both handheld RFID readers, which can be adjusted for different reading ranges, or fixed readers with wider coverage for real time inventory counts, but less precise localization. To prevent errors and false reads (that is reading a tag in the store area from an inventory count carried out in the backroom and vice versa), both handheld and fixed readers need physical shielding materials, like metal foils on store walls, to contain the RFID signals within specific areas, typically backroom area and sales floor area. However, this approach has drawbacks, including cost, limited flexibility, low scalability, and aesthetic concerns, especially in open sales floor layouts. A recent study by G. Esposito, Mezzogori, Neroni, Rizzi, and Romagnoli introduced a software-based shielding approach that eliminates the need for physical barriers between backrooms and store area. This method uses item-level tags and a machine learning algorithms to estimate tag locations without physical shielding. The SBS approach is more cost-effective and flexible, with an average accuracy of around 95%. However, it was tested in a specific scenario with plasterboard walls and fixed room sizes. This study aims to assess various methods for indoor item localization using passive RFID tags, addressing stationary and moving objects' detection and tracking. It also evaluates the Software-Based Shielding approach with different wall types, thickness, tag arrangements, and densities. The findings related to this activity were published in a dedicated paper by Neroni, Rizzi, Romagnoli, and Rosa. The performance of Software-Based Shielding was further tested in two fashion retail stores in Bologna and Milan, Italy, While the brand name is kept confidential, these tests evaluate the approach's accuracy in real retail environments considering factors like reader models, power levels, and classification methods. The results pertaining to this topic were published in a dedicated paper by Mezzogori, Rizzi, Romagnoli, and Rosa. The study also aims at implementing the SBS algorithm into the id-Bridge RFID platform suite developed by Murata ID Solutions. In conclusion, this research demonstrates that Software-Based Shielding is a practical solution for improving inventory count processes in fashion retail stores, avoiding the need of physical barriers between store backroom and sales floor. It saves time, ensures accurate item tracking, and is adaptable to various environments, making it a valuable tool for broader adoption. Future work will be aimed at assessing the performance of the implemented algorithm in different real case applications, in order to better understand its practical scalability, implications and performances.