METHODOLOGY FOR SHELF PRODUCT RECOGNITION AND INTELLIGENT REPLENISHMENT DECISION-MAKING USING YOLOV11
Keywords:
Shelf product recognition, Intelligent replenishment decisions, Object detection and counting, Demand forecasting, Inventory managementAbstract
To address the issues of delayed inventory visibility and experience-driven retail shelf management, this paper proposes an intelligent replenishment decision-making framework that integrates computer vision and management science. First, a model for shelf item recognition and quantity counting is constructed using You Only Look Once version 11 (YOLOv11), enabling real-time inventory sensing. On this basis, inventory control theory is incorporated to establish a dynamic safety stock warning mechanism. Combined with exponential smoothing for demand forecasting, the dynamic replenishment quantity is calculated as “Replenishment Quantity = Forecasted Demand – Current Inventory + Safety Stock.” Using a self-built shelf image dataset, this paper compares YOLOv11 models of various scales, among which the m model achieves the optimal balance between computational efficiency and performance, with an average product detection accuracy (mAP50) of 60.2%; Compared to other mainstream detection models (Faster R-CNN, RetinaNet, YOLOv8m), YOLOv11m achieved the highest accuracy with the fewest parameters, validating the reliability of the visual perception module. Theoretical analysis, case simulations, and ablation experiments demonstrate that the proposed dynamic replenishment model can more effectively address demand fluctuations and reduce the risks of stockouts and inventory buildup compared to fixed-threshold strategies and minimum-maximum inventory strategies; furthermore, safety stock and demand forecasting are complementary and indispensable. This study provides a feasible solution and theoretical reference for visual inventory management and scientific replenishment decision-making in smart retail.References
[1] Gruen T W, Corsten D S, Bharadwaj S. Retail out-of-stocks: A worldwide examination of extent, causes and consumer responses. Washington, DC: Grocery Manufacturers of America, 2002.
[2] Jiang Honghua. Several Suggestions for Improving the Supermarket Inventory Counting System. Times Finance, 2014 (4Z): 173–174.
[3] Kikuchi T, Takeuchi S. Self-supervised human-object interaction of complex scenes with context-aware mixing: Towards in-store consumer behavior analysis//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024: 744-751.
[4] Varol G, Kuzu R S. Toward retail product recognition on grocery shelves//Sixth International Conference on Graphic and Image Processing (ICGIP 2014), SPIE, 2015, 9443: 46-52.
[5] Sun Haixia. Study on segmentation and recognition methods of supermarket shelf. Master’s thesis, Nanjing University of Technology and Engineering, 2014.
[6] Goldman E, Herzig R, Eisenschtat A, et al. Precise detection in densely packed scenes//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019: 5227-5236.
[7] Guo Z, Wang G, Xie K. A lightweight YOLOv11-based commodity detector for dense smart retail cabinets. Scientific Reports, 2026.
[8] Chen B, Li L, Wang Y, et al. RetailDet: An Efficient Fusion Attention Network for Joint Product and Vacancy Identification in Smart Retail. IEEE Access, 2025.
[9] Pietrini R, Paolanti M, Mancini A, et al. Shelf Management: A deep learning-based system for shelf visual monitoring. Expert Systems with Applications, 2024, 255: 124635.
[10] Alghaslan M, Almutairy K, El-Alfy E S, et al. RetailEye: Supervised Contrastive Learning with Compliance Matching for Retail Shelf Monitoring//International Conference on Neural Information Processing, Singapore: Springer Nature Singapore, 2024: 248-263.