Analisis Pengolahan Citra Produk Pangan Olahan Kemasan Untuk Deteksi Kehalalan Menggunakan Metode Optical Character Recognition (OCR) dan Klasifikasi Support Vector Machine (SVM) serta Convolutional Neural Network (CNN)

Abstract

This research aims to detect the halal status of packaged processed food products with OCR technology and machine learning based classification methods. Datasets in the form of product packaging images are collected through direct documentation and official sources, then processed using image preprocessing (grayscale, Otsu thresholding, resize 2x) to improve text reading quality. The OCR extracted text is then processed through the cleaning stage (lowercasing, normalisation, stopword removal) and analysed using a spaCy based NER approach with non-halal ingredient entity rules. The labelled data is used as input for two classification models SVM with TF-IDF representation and CNN with token-sequence representation. Evaluation was conducted using accuracy, precision, recall, F1-score, and confusion matrix metrics on 80:20 and 70:30 data split scenarios. The results showed that the SVM model recorded the highest accuracy of 99.36% and macro F1-score of 97.05% in the 80:20 scenario, and 99.22% and 96.29% in the 70:30 scenario. The CNN model recorded an accuracy of 98.61% and macro F1-score of 93.78% in the 80:20 scenario, as well as 98.08% and 91.50% in the 70:30 scenario. SVM model shows stability and consistently higher accuracy, while CNN remains competitive mainly due to its ability to capture contextual patterns from text data through embedding and convolution.