AgriFreshNET Freshness and Shelf-Life Image Dataset
Description
The AgriFreshNET Dataset is a large-scale image dataset developed for automated produce freshness assessment and shelf-life prediction. It contains 14,500 RGB images across 24 classes of fruits and vegetables, with samples categorized into Fresh, Semi-Fresh, and Rotten stages. Images were collected from real-world environments, including local markets, supermarkets, households, farms, and storage facilities, using smartphone cameras under diverse lighting conditions, backgrounds, viewing angles, and object orientations. The dataset is relatively balanced, with an average of 569 images per class, making it suitable for supervised learning. In addition to categorical freshness labels, each sample includes an approximate post-harvest shelf-life annotation (e.g., 1–4, 4–7, 8–12, and 24–35 days, depending on the produce type). These annotations enable models to learn gradual spoilage patterns and estimate remaining shelf life more accurately. AgriFreshNET supports a wide range of computer vision applications, including freshness classification, quality assessment, shelf-life prediction, transfer learning, explainable AI, and smart agriculture. Its diverse acquisition conditions and realistic annotations make it a valuable benchmark for developing robust AI models for food quality monitoring and reducing post-harvest food waste.
Files
Steps to reproduce
Download the Raw Data.zip file containing the original RGB images collected from multiple real-world environments. Use the Processed_Data.zip file if you want to directly train and evaluate machine learning or deep learning models without performing preprocessing. If starting from the raw dataset, resize the images, normalize pixel values, verify labels, and optionally apply data augmentation (e.g., rotation, flipping, zooming, and brightness adjustment). Organize the processed images according to the provided class and freshness labels.
Institutions
- East West UniversityDhaka Division, Dhaka