| Household food waste that is a significant contributor to global environmental and economic challenge often stems from consumers' difficulty in accurately assessing produce ripeness. This paper proposes a two-stage deep learning system for smart refrigerators to address this challenge. The system first identifies common fruits using an object detection model and subsequently classifies tomato ripeness into four physiological stages. To rigorously evaluate the performance of YOLOv11 and Faster R-CNN architectures, we developed two custom datasets capturing eight common fruit types and the full ripening lifecycle of tomatoes under controlled conditions. Experimental results demonstrate that the YOLOv11-based system achieves superior performance, maintaining over 95% accuracy in both stages with a low end-to-end latency of 9.1s. This solution offers a practical, real-time tool for intelligent home appliances to actively assist consumers in reducing food waste. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.