Otomatisasi Pembacaan Alat Pengukur Analog Berbasis YOLOv9, Segment Anything Model, dan TR-OCR

Asmawati, Diah (2026) Otomatisasi Pembacaan Alat Pengukur Analog Berbasis YOLOv9, Segment Anything Model, dan TR-OCR. Masters thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 6025232014-Master_Thesis.pdf] Text
6025232014-Master_Thesis.pdf - Accepted Version
Restricted to Repository staff only

Download (4MB) | Request a copy

Abstract

Penelitian ini bertujuan untuk menjalankan otomatisasi pada pembacaan alat pengukur analog yang masih umum digunakan di berbagai sektor industri. Tantangan utama yang dihadapi meliputi permasalahan beragamnya jenis alat pengukur, gangguan pencahayaan, citra yang tidak jernih atau sudut pandang yang miring. Metode yang diusulkan dapat meningkatkan efisiensi, akurasi, dan keselamatan kerja dengan mengeliminasi keterlibatan manusia secara langsung. Sistem yang dirancang dalam penelitian ini menggunakan pendekatan pembelajaran mendalam yang menggabungkan model-model yang spesifik. Pendekatan ini melibatkan metode YOLOv9 untuk mendeteksi alat dan komponen instrumen. Hasil deteksi menjadi masukan prompt bagi proses segmentasi jarum dan garis-garis melalui model Segment Anything Model (SAM) yang telah dilatih khusus. Susunan garis-garis skala membentuk elips busur skala yang merepresentasikan nilai-nilai pengukuran berdasarkan sudut jarum terhadap pusat skala. Teknik ini menjawab tantangan praproses yang melibatkan koreksi perspektif penampang alat seperti pada studi-studi sebelumnya. Sistem juga menerapkan TrOCR untuk pengenalan angka dan menerjemahkan penunjuk jarum menjadi nilai numerik. Dataset penelitian terdiri dari kumpulan citra alat pengukur analog dari berbagai sumber, termasuk dataset publik MC1296, yang mencakup berbagai tipe alat pengukur, sudut pandang, dan kondisi visual. Hasil evaluasi menunjukkan bahwa sistem memiliki tingkat akurasi yang tinggi, dengan Average Relative Error (ARelE) sebesar 1,935% dan Average Reference Error (ARefE) sebesar 0,601% terhadap skala alat ukur. Secara keseluruhan, pendekatan yang dikembangkan terbukti efektif dan dapat diandalkan dalam berbagai kondisi citra alat pengukur.
============================================================================================================================
This research aims to automate the reading of analog gauges, which remain widely used across various industrial sectors. The primary challenges to address include issues with different measuring instruments, lighting disturbances, unclear images, and oblique viewing angles. The proposed method can improve efficiency, accuracy, and work safety by eliminating direct human involvement. The system adopts a deep learning-based approach that integrates multiple specialized models. YOLOv9 is employed to detect instruments and their components. The detection outputs then serve as prompts for segmentation of needles and scale ticks (both main and secondary ticks) using a fine-tuned Segment Anything Model (SAM). The arrangement of detailed ticks forms an elliptical arc, representing the measurement scale, from which the needle’s angular position relative to the scale’s center is determined. This formulation also addresses the preprocessing challenges in perspective adjustment encountered in previous studies. TrOCR is then applied for numerical character recognition, enabling the translation of the needle’s position into precise measurement values. Our research utilizes a dataset comprising images of analog gauges from various sources, including the publicly available MC1296 dataset, encompassing multiple instrument types, viewing angles, and visual conditions. Experimental results demonstrate that the system achieves high accuracy, with an Average Relative Error (ARelE) of 1,935% and an Average Reference Error (ARefE) of 0,601%. Overall, the proposed approach proved effective and reliable under diverse imaging conditions, offering a robust solution for automating analog measurement reading in industrial applications.

Item Type: Thesis (Masters)
Uncontrolled Keywords: alat pengukur analog, pembelajaran mendalam, SAM, TR-OCR, YOLOv9,analog gauge, deep learning, SAM, TR-OCR, YOLOv9
Subjects: T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User: Diah Asmawati
Date Deposited: 30 Jan 2026 02:49
Last Modified: 30 Jan 2026 02:49
URI: http://repository.its.ac.id/id/eprint/131194

Actions (login required)

View Item View Item