Thermal-Images-of-Bread-Contamination-Labeled

This dataset consists of thermal images of bread slices, both contaminated and uncontaminated, designed for training and evaluating computer vision models for contamination detection. Commercial white bread was used, with slices standardized for size (10x10x1 cm) and moisture content (35 ± 2%). Four common food processing contaminants (deionized water, refined vegetable oil, food-grade machinery grease, and a diluted cleaning agent) were applied at two different levels, with 50 bread slices prepared for each contaminant and 50 uncontaminated controls. Contaminant application was carefully controlled using micropipettes and spatulas, with triplicate runs for each scenario. Thermal images were captured using a FLIR One Pro camera (160x120 resolution, 70 mK sensitivity) within a custom-built, light- and temperature-controlled enclosure. A hot-air blower provided a heat stimulus (35°C airflow for 2 or 3 seconds) at a 45° angle, and a conveyor system with a non-reflective black surface ensured consistent imaging. A total of 650 thermal images were acquired. Preprocessing and augmentation were performed using Roboflow software. Preprocessing steps included auto-orientation, cropping, resizing to 640x640, auto-contrast adjustment, and grayscale conversion. Augmentation techniques included flipping and exposure adjustments. The final dataset comprises 1342 training images, 56 validation images, and 28 test images along with their labels.