A Thermal Imaging Dataset for Fruit Bruise Detection and Classification
Description
This dataset contains thermal images of fruits for automated bruise detection and classification tasks using deep learning methods. The dataset includes two fruit types: apple and tomato, each categorized into bruised and healthy classes. The dataset is organized into a hierarchical folder structure as follows: - apple/ - raw/ - bruised/ - healthy/ - processed/ - bruised/ - healthy/ - tomato/ - raw/ - bruised/ - healthy/ - processed/ - bruised/ - healthy/ The total number of images is as follows: - Apple (Bruised): 355 images - Apple (Healthy): 271 images - Tomato (Bruised): 195 images - Tomato (Healthy): 195 images The dataset is provided without predefined training, validation, or test splits to allow researchers flexibility in designing their own experimental protocols. These thermal images are suitable for tasks such as: - Binary classification (healthy vs. bruised) - Cross-dataset generalization - Transfer learning and domain adaptation studies - Explainable AI (XAI) analysis using thermal patterns If you use this dataset in your research, please cite the following article: Nogay, H. S., & Nogay, N. H. (2026). Thermal imaging-based characterization of fruit bruising using deep learning: a cross-dataset evaluation study. European Food Research and Technology, 252, 394. DOI: 10.1007/s00217-026-05324-0
Files
Institutions
- Bursa Uludağ Üni̇versi̇tesi̇Bursa Province, Bursa