EVALUASI KEPUASAN PELANGGAN BERDASARKAN EKSPRESI WAJAH MENGGUNAKAN REAL TIME DETECTION TRANSFORMER (RT-DETR)

Shafa Meira Wahyono, - (2025) EVALUASI KEPUASAN PELANGGAN BERDASARKAN EKSPRESI WAJAH MENGGUNAKAN REAL TIME DETECTION TRANSFORMER (RT-DETR). S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Ekspresi wajah merupakan indikator paling baik dalam mengetahui perasaan manusia karena hanya teridentifikasi dalam waktu singkat yaitu 0,5 detik. Maka dari itu, ekspresi wajah dapat digunakan sebagai indikator evaluasi kepuasan pelanggan. Namun, karena perubahan ekspresi wajah yang cepat, penggunaan teknologi untuk pengenalan wajah serta klasifikasi ekspresi menjadi lebih ideal. Selain itu, pengimplementasian teknologi di dunia nyata sering kali dihadapi tantangan, seperti keterbatasan perangkat keras. Berdasarkan permasalahan tersebut, dilakukan penelitian untuk membangun model kecerdasan buatan yang mampu mengevaluasi kepuasan pelanggan berdasarkan pengenalan ekspresi wajah dengan memerhatikan kemungkinan keterbatasan sumber daya yang dapat muncul ketika pengimplementasian. Pertama, model dikembangkan untuk mendeteksi wajah menggunakan metode RT-DETR (Real Time-Detection Transformer) dengan backbone ResNet-18 dan LCNet-0.25 (Lightweight CPU Convolutional Neural Network). Kedua, model mengklasifikasikan 2 tipe ekspresi wajah, yaitu 7 ekspresi wajah dan 3 ekspresi wajah, menggunakan metode Real-Time CNN. Penelitian yang dilakukan, menunjukan bahwa metode RT-DETR dengan backbone ResNet-18 mendapatkan kinerja terbaik dengan 35.0% AP, 6.64 FPS, dan parameter 20M serta Real-Time CNN dengan 3 ekspresi wajah dengan performa micro-average F1-score 68.4%. Kombinasi dari model RT-DETR dan model Real-Time CNN memiliki kinerja 2.1% AP dan 4.7 FPS. Facial expression is the best indicator to recognize emotion because it is only recognizable in a such short time, which is about 0,5 seconds. Therefore, facial expression is usable as customer satisfaction evaluation indicator. However, as facial expression changes quickly, it would be more ideal to use technology to recognize the faces and classify the expression. Furthermore, technology implementation in real-world was often meet with computation limitation. Based on those problems, this research is done to build Artificial Intelligence model that can evaluate customer satisfaction through facial expression, while keeping attention on the possibility of resource limitation when implementing. Firstly, a model is developed to detect faces using methods, such as RT-DETR (Real Time-Detection Transformer) with backbone ResNet-18 and LCNet-0.25 (Lightweight CPU Convolutional Neural Network. Secondly, a model will classify 2 types of facial expression, such as 7 type of facial expression and 3 type of facial expression, using Real-Time CNN method. RT-DETR with ResNet-18 as backbone achieves best performance with 35.0% AP, 6.64 FPS, and parameter of 20M, while Real-Time CNN with 3 types facial expression achieves micro-average F1-score of 68.4%. The combination of RT-DETR and Real Time-CNN achieves 4.7% AP dan 5.94 FPS.

	Text S_KOM_2007723_Title.pdf Download (639kB)
	Text S_KOM_2007723_Chapter1.pdf Download (125kB)
	Text S_KOM_2007723_Chapter2.pdf Restricted to Staf Perpustakaan Download (1MB) | Request a copy
	Text S_KOM_2007723_Chapter3.pdf Download (421kB)
	Text S_KOM_2007723_Chapter4.pdf Restricted to Staf Perpustakaan Download (1MB) | Request a copy
	Text S_KOM_2007723_Chapter5.pdf Download (43kB)

Item Type:	Thesis (S1)
Additional Information:	ID SINTA Dosen Pembimbing: Munir: 0025036602 Yaya Wihardi: 0025038901
Uncontrolled Keywords:	Detection Transformer, Pengenalan Ekspresi Wajah, Real Time Customer Satisfaction, Detection Transformer, Facial Expression Recognition, Real Time
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Divisions:	Fakultas Pendidikan Matematika dan Ilmu Pengetahuan Alam > Program Studi Ilmu Komputer
Depositing User:	Shafa Meira Wahyono
Date Deposited:	20 Feb 2025 02:48
Last Modified:	20 Feb 2025 02:48
URI:	http://repository.upi.edu/id/eprint/130840

Actions (login required)

View Item