Verifikasi Kode Produksi Menggunakan Teknologi Optical Character Recognition(OCR) Dengan Metode Single Visual Model For Scene Text Recognition(SVTR)

Atmosukarto, Dhevand Cody (2025) Verifikasi Kode Produksi Menggunakan Teknologi Optical Character Recognition(OCR) Dengan Metode Single Visual Model For Scene Text Recognition(SVTR). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040211092_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Kesalahan pengetikan kode produksi pada layar jet-print di lini pengemasan PT Charoen Pokphand Indonesia berpotensi menimbulkan pemrosesan ulang, penarikan produk, dan penurunan kepercayaan konsumen, padahal kode produksi kombinasi huruf dan angka yang diwajibkan Peraturan BPOM No. 31/2018 menjamin keterlacakan tanggal produksi, nomor batch, serta umur simpan. Dalam satu kasus yang pernah terjadi, kesalahan input kode berlangsung selama satu hari penuh dan menyebabkan 13 batch produksi tercetak dengan kode yang salah, yang berdampak pada penarikan ulang produk dan keterlambatan distribusi. Untuk mengatasi risiko ini tanpa memodifikasi mesin vendor-owned tertutup, studi ini mengembangkan verifikator otomatis berbasis Optical Character Recognition (OCR) dengan arsitektur Single Visual Model for Scene Text Recognition (SVTR). Kebaruan penelitian terletak pada evaluasi sistematis tiga Learning Rate 0,001, 0,0005, dan 0,00025 dan lima komposisi Dataset (Lapangan, ICDAR15, Sintetik, Lapangan+ICDAR15, Lapangan+Sintetik). Model SVTR hasil fine-tuning dibandingkan baseline PPOCRv3 memakai metrik Word Error Rate (WER), Character Error Rate (CER), serta precision, recall, dan F1-score per karakter. Konfigurasi Learning Rate 0,00025 pada gabungan Lapangan+Sintetik mencapai WER 0,4098 dan CER 0,1637 mengungguli baseline (WER 0,6976 dan CER 0,2722) serta dua Learning Rate lebih besar dengan peningkatan akurasi paling jelas pada karakter minor dan bentuk kompleks. Temuan ini menegaskan bahwa konfigurasi SVTR dengan fine-tuning terarah dan komposisi data Lapangan+Sintetik mampu menghasilkan sistem verifikasi kode produksi yang cepat rata-rata 5,13 detik serta memiliki nilai error yang rendah (CER 0,1637 dan WER 0,4098), menjadikannya solusi praktis tanpa perlu modifikasi perangkat keras lini produksi.
=====================================================================================================================================
Typing errors in production codes on the jet-print interface at the packaging line of PT Charoen Pokphand Indonesia pose significant risks of reprocessing, product recalls, and diminished consumer trust. These codes, which consist of alphanumeric combinations mandated by BPOM Regulation No. 31/2018, are essential for ensuring traceability of production dates, batch numbers, and shelf life. In one recorded case, a mistyped production code persisted for an entire day, resulting in 13 production batches being printed with incorrect codes, which led to product recalls and distribution delays. To mitigate such risks without modifying the closed vendor-owned machines, this study developed an automated verification system based on Optical Character Recognition (OCR) using the Single Visual Model for Scene Text Recognition (SVTR) architecture. The novelty of this research lies in the systematic evaluation of three Learning Rates (0.001, 0.0005, and 0.00025) and five Dataset compositions (Field, ICDAR15, Synthetic, Field+ICDAR15, Field+Synthetic). The fine-tuned SVTR model was compared with the PPOCRv3 baseline using Word Error Rate (WER), Character Error Rate (CER), and per-character precision, recall, and F1-score as evaluation metrics. The configuration using a Learning Rate of 0.00025 on the Field+Synthetic Dataset achieved a WER of 0.4098 and a CER of 0.1637, outperforming the baseline (WER 0.6976 and CER 0.2722) and the two higher Learning Rates, with the most notable accuracy improvements observed in minor and complex character forms. These findings confirm that the SVTR configuration, with targeted fine-tuning and a Field+Synthetic Dataset composition, yields a fast verification system averaging 5.13 seconds per instance and high precision (CER 0.1637; WER 0.4098), making it a practical solution deployable without hardware modification to the production line.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Optical Character Recognition (OCR), Kode Produksi, Single Visual Model for Sccene Text Recognition(SVTR), Learning Rate, Optical Character Recognition (OCR), Production Code, Single Visual Model for Sccene Text Recognition(SVTR), Learning Rate.
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Dhevand Cody Atmosukarto
Date Deposited:	04 Aug 2025 09:33
Last Modified:	04 Aug 2025 09:33
URI:	http://repository.its.ac.id/id/eprint/124490

Actions (login required)

View Item