Detection and Classification of Plastic Contaminants on Chicken Meat using Color and Hyperspectral Imaging

This paper proposes an innovative two-step deep learning methodology for detecting and classifying various small plastic contaminants (5 mm × 5 mm and 2 mm × 2 mm) on chicken breast fillets. It combines YOLO object detection with RGB images for precise localization and 1D-CNN classification with hyperspectral images (600-1,700 nm) for accurate material identification, utilizing 12 common polymer types like HDPE, LDPE, PET, PVC, and PP. The study employed 52 color and 52 hyperspectral images. Both YOLOv7x and YOLOv8x excelled at detecting 5 mm plastic foreign materials (FMs) with a detection accuracy of 99 %. However, YOLOv7x outperformed YOLOv8x by 15 % for 2 mm FMs (80% vs. 65%). The classification accuracy of the 1D-CNN model using hyperspectral images was 97 % in classifying 12 common plastic contaminant types on chicken meat surfaces. This study demonstrates color imaging facilitating real-time detection of foreign materials and hyperspectral imaging enabling precise material identification. The study results pave the way for efficient and automated quality control and enhancement of safety standards in poultry processing.