RESUMO
Stockouts constitute a major challenge in the retail industry. Stockouts are caused by errors related to manual stockkeeping and by the misplacement of items on shelves. Such errors account for up to 4% of lost sales. Real-time inventory management systems for misplaced items or missing stock detection in retail stores are limited. Accordingly, a conductive polymer-based interactive shelving system for real-time inventory management is developed. The system comprises an 80 × 48 sensor array fabricated by screen-printing a piezoresistive carbon-based conductive polymer layer onto gold interdigitated electrodes deposited on a flexible substrate. Each sensing pixel has dimensions of 5 mm × 5 mm and a sensing area of 4 mm × 4 mm. The sensor mat can detect the shape and weight features of stockkeeping units (SKUs), which can then be analyzed by a TensorFlow model for SKU identification. The developed system is characterized for functional resistance range, uniformity, repeatability, and durability. The accuracy of SKU identification achieved using shape features only and the accuracy of SKU identification achieved using both shape and weight features is 95% and 99.2%, respectively. The key novelty of the work is the development of a deep learning-embedded interactive smart shelving system for retail inventory management by using the shape and weight features of SKU. Also, the developed system helps to detect the SKU if they are stacked one over the other. Furthermore, multiple sensor mats implemented on various shelves in a retail store can be modularized and integrated for monitoring under the control of a single PC. Accordingly, the proposed retail inventory tracking system can facilitate the development of automated "humanless" shops.
