Research Article
Detection of Freshness of Fruits and Vegetables for Food Delivery and Supply Chain Applications Using Deep Learning
@INPROCEEDINGS{10.4108/eai.19-12-2025.2366215, author={Nikita S. Pande and Prasad Kannawar and Venkatesh Mahindra and Ananya Peddamgari and B. Shiva Ganesh}, title={Detection of Freshness of Fruits and Vegetables for Food Delivery and Supply Chain Applications Using Deep Learning }, proceedings={Proceedings of the 1st International Conference on Computational Techniques in Data Science, IICTDS 2025, 19-20 December 2025, Chandigarh, India}, publisher={EAI}, proceedings_a={IICTDS}, year={2026}, month={11}, keywords={EfficientNet-B0 Capsule Network Hybrid Convolutional--Liquid Neural Network (Hybrid CNN--LNN) Freshness Detection Deep Learning Computer Vision Food Delivery Supply Chain Automation Image Classification Agri-Tech}, doi={10.4108/eai.19-12-2025.2366215} }- Nikita S. Pande
Prasad Kannawar
Venkatesh Mahindra
Ananya Peddamgari
B. Shiva Ganesh
Year: 2026
Detection of Freshness of Fruits and Vegetables for Food Delivery and Supply Chain Applications Using Deep Learning
IICTDS
EAI
DOI: 10.4108/eai.19-12-2025.2366215
Abstract
Food delivery, supply chain, and online grocery platforms are the main users of fresh fruits and vegetables. But the quality check is still done manually, which becomes slow and inconsistent during peak hours. This paper is about the development a deep learning–based automatic freshness detection system that can deliver and supply chain operations. We have considered a dataset of 12,000 images for training our model for 20 classes, which is pre-processed for resizing, normalization, and augmentation. Three architectures - EfficientNet-B0, Capsule Network (CapsNet), and Hybrid CNN–LNN - were trained with the same PyTorch–CUDA configurations. EfficientNet-B0 outperformed CapsNet and Hybrid CNN–LNNwith the highest accuracy (97.37%) as compared to CapsNet (84.96%) Hybrid CNN–LNN (72.25%). Efficient Net-B0 provided the best accuracy–efficiency trade-off, whereas CapsNet could be used for spatial interpretability and Hybrid CNN–LNN for edge devices adaptability. A Gradio-based interface enables real-time inference and freshness approval based on the threshold that helps packaging decisions, saves time, and ensures quality. The results show the promise of deep learning as a solution for automated freshness assessment that is scalable and can be integrated into modern food delivery ecosystems