Skripsi

DETEKSI EKSPRESI MATEMATIKA PADA CITRA DOKUMEN DIGITAL MENGGUNAKAN ARSITEKTUR DETECTION TRANSFORMER

---

Scientific digital documents often contain crucial information presented in the form of isolated mathematical expressions. Conventional Convolutional Neural Network (CNN)-based object detection methods, such as Faster R-CNN and YOLO, generally rely on manual components like anchor boxes and Non-Maximum Suppression (NMS) processes, which limit the model's flexibility regarding complex visual structure variations. This study aims to implement the Detection Transformer (DETR) architecture with a ResNet-50 backbone to detect isolated mathematical expressions in an end-to-end manner by formulating detection as a direct set prediction problem. Software development was conducted using the Rational Unified Process (RUP) method, covering the Inception, Elaboration, Construction, and Transition phases. The dataset used consists of 5,975 digital document images from IBEM-Va01. Based on the test results, the best model performance was achieved at 100 epochs, yielding a Mean Average Precision (mAP) at IoU 0.50 of 0.951 and an mAP in the IoU range of 0.50–0.95 of 0.649. These results indicate that the DETR architecture is capable of detecting and localizing mathematical expressions with high precision without requiring hand-crafted components utilized in previous methods.

---

Availability

Detail Information

Series Title

-

Call Number

T1934092025

Publisher

Indralaya : Prodi Kesehatan Masyarakat, Fakultas Kesehatan Masyarakat Universitas Sriwijaya., 2026

Collation

xvii, V-4 hlm.; ilus.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

006.420 7

Content Type

Text

Media Type

unmediated

Carrier Type

-

Edition

-

Subject(s)

Prodi Teknik Informatika
Detection Transformer (DETR)

Specific Detail Info

-

Statement of Responsibility

KA

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment