From Surface to Signature: Vision-Based Identification of Plastic Parts via Fluorescent Particle Patterns

This research proposes a data-driven method for unique, mark-free identification of plastic components based on stochastic fluorescent particle distributions. Addressing limitations of conventional traceability (T&T) systems, such as dependence on labels, barcodes, or RFID that are prone to wear or counterfeiting. This approach enables inherent physical identification without additional tags. By embedding fluorescent additives into various thermoplastics (PE, PP, PA12, PPA) via compounding and injection molding, each part exhibits a random surface pattern when illuminated under UV light (365 nm). These patterns serve as physical identifiers, enabling traceability without conventional markings such as labels or barcodes. A custom imaging setup was developed to capture these patterns, and a computer vision pipeline was implemented using blob detection to extract spatial features. From these, compact descriptor vectors were computed to enable similarity-based matching and classification. Compared to prior work on Physical Unclonable Functions (PUFs) and fluorescent additives, this method achieves significant data compression and demonstrates robustness against industrial surface wear and manufacturing variation, ensuring consistent identification across trials. Material color was found to influence detection quality, requiring adaptive calibration of additive concentrations. This work highlights the potential of combining material science and machine learning to enable non-invasive, counterfeit-resistant traceability in manufacturing environments. It provides a foundation for further research in physical instance recognition, image-based object fingerprinting, and data-driven part authentication.