Skripsi

SISTEM PENGENALAN WAJAH BERBASIS DEEP LEARNING PADA SISTEM KEAMANAN UNTUK AKSES MASUK

---

Developing a reliable and stable security system using biometric face recognition is a quite significant challenge, mainly due to the complexity and diversity of the human facial population state. This study aims to overcome this limitations of conventional security systems and stability issues found in some previous methods utilizing Transfer Learning. The core focus is to implement a modern security system for room access by developing and training Convolutional Neural Network (CNN) architectures independently from scratch. The method employed is Deep Learning, comparing two major architectures: ResNet-50 and VGG-16, in the process of real-time face detection and recognition. The training process was carried out for 25 Epochs and optimized using the Automatic Mixed Precision (AMP) technique for time and memory efficiency. The research results indicate that both models achieved higher performance and proved to be a Good Fit during training. In the test data evaluation, ResNet-50 showed quantitatively superior results with a accuracy of 99.15% and a loss of 0.03%, and a perfect Confusion Matrix with zero misclassifications. VGG-16 also yielded a high accuracy of 97.86% and loss 0.06%. The real-time face recognition are tested on 5 participants showed that ResNet-50 Model can achieve higher accuracy for face recognition compared to VGG-16. However, VGG-16 shows quite stable generalization (>95%) across several facial data points, indicating more robustness on some data. Overall, this research successfully implemented an accurate and consistent access control system, where the model's classification results were translated into physical signals to open or lock a door, proving the effectiveness of the from- scratch CNN architecture development approach for modern security systems.

---

Availability

Detail Information

Series Title

-

Call Number

T1904712025

Publisher

Indralaya : Prodi Teknik Elektro, Fakultas Teknik Universitas Sriwijaya., 2025

Collation

xv, 82 hlm.; ilus.; tab.; 29 cm.

Language

Indonesia

ISBN/ISSN

-

Classification

621.389 280 7

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Prodi Teknik Elektro
Sistem Keamanan Elektronik

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment