Thermal imaging-guided detection of transparent plastic contaminants on chicken breast: A combined vision and simulation approach.

Poultry consumption has surged in recent years, raising food safety concerns from foreign material (FM) contamination, particularly transparent plastics that are difficult to detect using conventional methods like X-ray imaging. This study aims to develop a robust detection system for transparent plastic contaminants on chicken meat using thermal imaging integrated with machine learning. Transparent plastics commonly found in the processing plants were cut into various sizes (0.2-8 cm) and placed on raw breast fillets, generating 386 annotated thermal images. A lumped thermal resistance-capacitance (LTRC) model and COMSOL Multiphysics 3D transient heat transfer simulation were also employed to understand thermal behavior and internal heat retention of plastic patches. The thermal contrast between plastic and chicken reached up to 1.05 °C in the experiment setup. The YOLOv8s model achieved a detection precision of 0.983 and segmentation precision of 0.949. These results demonstrate the effectiveness of combining thermal imaging and deep learning to reliably detect low-density, transparent FMs, enhancing food safety and inspection capabilities in poultry processing.