<![CDATA[Checking product categories and expiry dates in grocery stores is still largely done manually, making the process error-prone and time-consuming, especially for packaging with small or inconsistently formatted date prints. This research builds a two-stage pipeline that automatically classifies grocery product categories and detects expiry dates from product images. The method consists of an image classification model based on transfer learning with a frozen ResNet50 backbone, trained on three product categories (Fruits, Dairy Products, and Bread), and an expiry-date detection module leveraging the CRAFT text detector built into EasyOCR, combined with regular-expression matching and a tiered OCR fallback mechanism. The classification model was trained on 398 training images and evaluated on 98 validation images, achieving a validation accuracy of 83.67%, well above the 33.3% random-guess baseline for three classes, with the highest F1-score on the Dairy Products category (0.91). Testing of the date-detection pipeline on 63 test images showed a detection precision of 70.0% and an overall valid-date rate of 44.4%, indicating that a pragmatic approach based on a ready-made text detector can function without training a dedicated object detector, although reliability still needs improvement in filtering false-positive date candidates. Evaluation of the integrated system further found that the initial design assumption, that OCR should only run on packaged categories, was not entirely accurate, since the Fruits category showed a higher success rate (83.3%) than the packaged category (40.4%). The main contribution of this research is a pragmatic approach that leverages a ready-made text detector to locate very small date objects without training a dedicated object detector from scratch, allowing the pipeline to be implemented end-to-end under limited data and computing resources.]]>
