Real-Time Expiry Date Verification of CIJ Dot-Matrix Codes on Glossy Food Packaging Using a Raspberry Pi-Based YOLOv8–OCR Pipeline

Abstract

Automatic verification of expiration dates in the food industry is critical for consumer safety, yet challenging due to the fragmented nature of Continuous Inkjet (CIJ) dot-matrix fonts and specular reflections on glossy packaging. This study proposes a cost-effective, embedded visual inspection system implemented on a Raspberry Pi 4 to address these issues. The system integrates YOLOv8-Nano for precise label localization and an application-specific integration of adaptive thresholding with a 3×3 dilation kernel, tailored to repair CIJ font structures under the constraints of embedded hardware prior to recognition by PaddleOCR. Experimental results demonstrate that this morphological approach significantly reduced the Character Error Rate (CER) from 18.4% (unprocessed) to 2.3%, effectively overcoming the segmentation limitations of standard Optical Character Recognition (OCR) engines. The system achieved a detection accuracy of 99.2%, specifically at a low conveyor speed (10 cm/s). Under varying environmental conditions, the system maintained 92.2% accuracy under normal lighting (300 Lux) and achieved 95.0% accuracy under glare (500 Lux), because the brighter illumination created greater localized contrast for the adaptive thresholding algorithm, making it easier to isolate the dark ink. With an inference latency of 229 ms (approx. 4.3 FPS), these findings confirm the system's viability as a robust, low-cost alternative to proprietary machine vision solutions for Small and Medium-sized Enterprises (SMEs) operating at moderate conveyor speeds.