IMPLEMENTASI ALGORITMA MTCNN DAN ARSITEKTUR VGG 16 UNTUK DETEKSI EMOSI MANUSIA BERDASARKAN EKSPRESI WAJAH

Dimas Setiawan, - (2022) IMPLEMENTASI ALGORITMA MTCNN DAN ARSITEKTUR VGG 16 UNTUK DETEKSI EMOSI MANUSIA BERDASARKAN EKSPRESI WAJAH. S1 thesis, Universitas Pendidikan Indonesia.

Preview	Text S_PSTI_1805720_Title.pdf Download (820kB) | Preview
Preview	Text S_PSTI_1805720_Chapter1.pdf Download (133kB) | Preview
	Text S_PSTI_1805720_Chapter2.pdf Restricted to Staf Perpustakaan Download (730kB)
Preview	Text S_PSTI_1805720_Chapter3.pdf Download (430kB) | Preview
	Text S_PSTI_1805720_Chapter4.pdf Restricted to Staf Perpustakaan Download (716kB)
Preview	Text S_PSTI_1805720_Chapter5.pdf Download (118kB) | Preview
	Text S_PSTI_1805720_Appendix.pdf Restricted to Staf Perpustakaan Download (1MB)

Abstract

Teknologi semakin maju membawa kebermanfaatan dan kabaharuan khususnya di Deep Learning, termasuk Computer Vision. Penerapan ini umumnya pada pembelajaran maupun karyawan di perusahaan, sulitnya mendeteksi ekspresi wajah dalam jumlah besar, lebih dari satu orang dalam kondisi yang sama. Penelitian ini berfokus pada implementasi Algoritma MTCNN dan Arsitektur VGG-16. Penelitian ini menggunakan Framework AI Project Life Cycle, penerapannya dengan framework streamlit dalam pengembangan sistemnya. Dataset yang digunakan FER 2013, terdiri dari 67.885 dataset, 7 jenis ekspresi yaitu angry, disgust, fear, happy, neutral, sad, dan surprise. Pembagian dataset dilakukan dengan membagi menjadi 3 yaitu Train, Validation, dan Testing, data train terdiri dari 42.825 gambar, data validation terdiri dari 10.704 gambar, dan data testing terdiri dari 14.356. Dalam proses training menghasilkan model terbaik dengan training accuracy mencapai 85,70 % dan testing accuracy mencapai 85,71 %, untuk training loss mencapai 1.7759 dan testing loss mencapai 1.7696. ROC AUC yang didapatkan stabil, tidak overfitting dengan ROC AUC score 94%. Sistem deteksi ini memiliki kelemahan jika pencahayaan dan wajah terpotong serta gelap, MTCNN tidak dapat mendeteksinya. ----- More advanced technology brings benefits and updates, especially in Deep Learning, including Computer Vision. This application is generally in learning and employees in the company, it is difficult to detect facial expressions in large numbers, more than one person in the same condition. This research focuses on the implementation of the MTCNN Algorithm and the VGG-16 Architecture. This research uses the AI Project Life Cycle Framework, its application with a streamlit framework in the development of the system. The dataset used by FER 2013, consists of 67,885 datasets, 7 types of expressions, namely angry, disgusted, fearful, happy, neutral, sad, and surprise. Distribution of the dataset is done by dividing into 3 namely Train, Validation, and Testing, train data consists of 42,825 images, data validation consists of 10,704 images, and testing data consists of 14,356. In the training process, it produces the best model with training accuracy reaching 85.70% and testing accuracy reaching 85.71%, for training losses reaching 1.7759 and testing losses reaching 1.7696. The ROC AUC obtained was stable, not overfitting with an ROC AUC score of 94%. This detection system has a weakness if the lighting and faces are cut off and dark, MTCNN cannot detect them.

Item Type:	Thesis (S1)
Additional Information:	Link Google Scholar Taufik Ridwan : https://scholar.google.com/citations?user=9RT7rk0AAAAJ&hl=en&oi=ao Suprih Widodo : https://scholar.google.com/citations?user=P-awTsUAAAAJ&hl=en&oi=ao ID SINTA Taufik Ridwan : 6757589 Suprih Widodo : 5978120
Uncontrolled Keywords:	AI Project Life Cycle, Algoritma MTCNN, Arsitektur VGG-16, Ekspresi Wajah
Subjects:	L Education > L Education (General) T Technology > T Technology (General)
Divisions:	UPI Kampus Purwakarta > S1 Pendidikan Sistem Teknologi dan Informasi
Depositing User:	Dimas Setiawan
Date Deposited:	08 Sep 2022 03:13
Last Modified:	08 Sep 2022 03:13
URI:	http://repository.upi.edu/id/eprint/78843

Actions (login required)

View Item