FFIS: Design and Development of a Domestic Fruit Freshness Recognition System
DOI:
https://doi.org/10.4108/eetpht.11.11042Keywords:
Digital health, Edge AI, Fruit freshness detection, Multi-task learning, On-device inference, Energy efficiencyAbstract
INTRODUCTION: In everyday households, fruits are often stored for too long, forgotten, or discarded early due to uncertainty about their freshness, leading to avoidable waste, unnecessary cost, and suboptimal dietary habits. A reliable and user-centered fresh produce assessment tool can help people establish a healthier and more sustainable lifestyle. The existing freshness assessment systems are mainly designed for industrial or laboratory environments and rarely meet the privacy, stability and cost requirements of family digital health products.
OBJECTIVES: This study aims to design and evaluate an edge-AI system that jointly recognizes fruit category and ordinal freshness stage in real kitchens, providing a reproducible benchmark for household-oriented freshness sensing rather than a one-off engineering prototype.
METHODS: We develop the Fruit Freshness Identification System (FFIS), a lightweight multi-task detector built on YOLO11n with a BiFPN neck, ACmix hybrid attention, and an IoU-based localization loss reweighted for partial occlusion. A kitchen-scene dataset is collected and stratified by household, countertop material, lighting, and clutter. The system is trained with an ordinal regression head for freshness staging and evaluated on COCO-style detection metrics, stage-wise classification metrics, and edge-device throughput and energy consumption.
RESULTS: On the FFIS-Fruit test split, FFIS achieves an mAP@0.5:0.95 of 62.4 ± 0.4, mAP@0.5 of 94.1 ± 0.3, and recall of 90.8 ± 0.3, outperforming lightweight YOLO baselines as well as a two-stage detector-plus-classifier pipeline. Warm-LED and glossy-countertop subsets show consistent gains in high-reflection scenarios. On-device experiments reach ~55 FPS (INT8) on Jetson Orin Nano and ~10 FPS on Raspberry Pi 5 under a unified evaluation protocol.
CONCLUSION: FFIS provides a household-oriented, privacy-preserving reference implementation for kitchen fruit monitoring, demonstrating that single-pass multi-task detection can meet domestic deployment constraints on low-cost hardware. Rather than directly claiming reductions in household food waste, the findings establish a technical foundation for integrating freshness signals into downstream digital-health workflows such as waste-aware reminders and dietary planning.
Downloads
References
[1] Forbes H. Food waste index report 2021.
[2] Yılmaz EN, Pehlivan ÖÜ. Using Household Fruit and Vegetable Waste in Recipes to Reduce Kitchen Food Waste and their Nutritional and Functional Values. Akademik Gıda.(Yeşil Dönüşüm Özel Sayısı):33-44.
[3] Xiayu LI. Research on Identification Methods for Healthy Diet Green Design. Green Design Engineering. 2024 Oct 1;1(1):12-9.
[4] Wang Y, Min J, Khuri J, Xue H. Digital health applications for improving nutrition and physical activity behaviors: A systematic review. Nutrients. 2020;12(2):335.
[5] Yousheng YA. Application of Inclusive Design in the Healthy Food Dashboard: Research on Diverse User Needs and Equity. Green Design Engineering. 2024 Oct 1;1(1):1-1.
[6] Jocher G, Chaurasia A, Stoken A, Borovec J, Kwon Y, Michael K, Fang J, Yifu Z, Wong C, Montes D, Wang Z. ultralytics/yolov5: v7. 0-yolov5 sota realtime instance segmentation. Zenodo. 2022 Nov 22.
[7] Jocher G, Chaurasia A, Qiu J. Ultralytics YOLO (v8.2.0) [Internet]. Zenodo; 2024 [cited 2025 Dec 13]. Available from: https://doi.org/10.5281/zenodo.10983461
[8] Jocher G, Qiu J, Chaurasia A. Ultralytics YOLO (v8.3.0) [Internet]. Zenodo; 2024 [cited 2025 Dec 13]. Available from: https://doi.org/10.5281/zenodo.13858602
[9] Sa I, Ge Z, Dayoub F, Upcroft B, Perez T, McCool C. DeepFruits: A fruit detection system using deep neural networks. Sensors (Basel). 2016;16(8):1222.doi:10.3390/ s16081222.
[10] Singh R, Nickhil C, Nisha R, Upendar K, Jithender B, et al. A comprehensive review of advanced deep learning approaches for food freshness detection. Food Eng Rev. 2025;17:127-160.doi:10.1007/s12393-024-09385-3.
[11] Shu Y, Zhang J, Wang Y, Wei Y. Fruit freshness classification and detection based on the ResNet-101 network and non-local attention mechanism. Foods. 2025;14(11):1987. doi:10.3390/foods14111987.
[12] Cao W, Mirjalili V, Raschka S. Rank consistent ordinal regression for neural networks with application to age estimation. Pattern Recognition Letters. 2020 Dec 1;140:325-31.
[13] Shi X, Cao W, Raschka S. Deep neural networks for rank-consistent ordinal regression based on conditional probabilities. Pattern Analysis and Applications. 2023 Aug; 26(3):941-55.
[14] Esfandiari Fard S, Ghosh T, Sazonov E. Multi-task NoisyViT for enhanced fruit and vegetable freshness detection and type classification. Sensors (Basel). 2025;25(19):5955. doi:10.3390/s25195955.
[15] Tan M, Pang R, Le QV. Efficientdet: Scalable and efficient object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 2020 (pp. 10781-10790).
[16] McCarthy B, Claesson A, Ghaffari M, et al. Promoting sustainable food behaviour with digital tools: A review. Curr Opin Food Sci. 2025;64:101328.doi:10.1016/j.cofs.2025.101328.
[17] Arshad R, Abdul-Malek Z, Parra-López C, Hassoun A, Qureshi MI, Sultan A, et al. Food loss and waste reduction by using industry 4.0 technologies: examples of promising strategies. Int J Food Sci Technol. 2025;60(1):vvaf034.doi:10.1093/ijfst/vvaf034.
[18] Pan X, Ge C, Lu R, Song S, Chen G, Huang Z, Huang G. On the integration of self-attention and convolution. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 2022 (pp. 815-825).
[19] Rezatofighi H, Tsoi N, Gwak J, et al. Generalized intersection over union: A metric and a loss for bounding box regression. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2019. p. 658–66.
[20] Zheng Z, Wang P, Liu W, Li J, Ye R, Ren D. Distance-IoU loss: Faster and better learning for bounding box regression. In Proceedings of the AAAI conference on artificial intelligence 2020 Apr 3 (Vol. 34, No. 07, pp. 12993-13000).
[21] Lin TY, Maire M, Belongie S, Hays J, Perona P, Ramanan D, Dollár P, Zitnick CL. Microsoft coco: Common objects in context. InEuropean conference on computer vision 2014 Sep 6 (pp. 740-755). Cham: Springer International Publishing.
[22] Efron B, Tibshirani RJ. An introduction to the bootstrap. New York: Chapman & Hall/CRC; 1993.
[23] Good P. Permutation, parametric and bootstrap tests of hypotheses. New York, NY: Springer New York; 2005 Dec 19.
[24] NVIDIA. TensorRT: High-performance deep learning inference SDK [Internet]. 2016–2025 [cited 2025 Jan 10]. Available from: https://docs.nvidia.com/tensorrt/
[25] Shridhar A, Tomson P, Innes M. Interoperating Deep Learning models with ONNX.jl. Proc JuliaCon Conf. 2020;1(1):59. doi:10.21105/jcon.00059.ONNX Runtime developers.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Xiayu Liu, Zhu Ke, Yuanhao Ding, Yingshan Tang, Yuqian Song
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
This is an open access article distributed under the terms of the CC BY-NC-SA 4.0, which permits copying, redistributing, remixing, transformation, and building upon the material in any medium so long as the original work is properly cited.
