Purba, Geoffrey Edmund Pierre and Ir. Satrio Hadi Wijoyo,, S.Si., S.Pd., and Ir. Nanang Yudi Setiawan,, S.T., (2024) Pengaruh Transfer Learning Resnet Dan Densenet Terhadap Performa Klasifikasi Ekspresi Wajah Menggunakan Dataset Fer-2013. Sarjana thesis, Universitas Brawijaya.
Abstract
Pengenalan ekspresi wajah adalah aspek penting dalam komunikasi nonverbal, dan pentingnya telah memicu penelitian yang signifikan dalam bidang kecerdasan buatan. Studi ini menyelidiki efek transfer learning menggunakan arsitektur Convolutional Neural Network (CNN), khususnya ResNet50 dan DenseNet121, pada performa pengenalan ekspresi wajah. Studi ini bertujuan untuk menganalisis dampak transfer learning pada performa mereka. Hasilnya menunjukkan bahwa implementasi transfer learning pada DenseNet121 dan ResNet50 dapat meningkatkan performa mereka secara signifikan, dengan DenseNet121 mencapai akurasi 95,1% pada set pelatihan dan 88.57% pada set validasi dan ResNet50 mencapai akurasi 100% pada set pelatihan dan 81.43% pada set validasi. Studi ini juga melakukan pengujian dengan melakukan prediksi menggunakan dataset baru yang tidak termasuk kedalam pelatihan model. Hasil pengujian menemukan bahwa fine-tuning model pretrained dapat meningkatkan performa model. Model DenseNet121 yang telah dilatih menggunakan FER-2013 dan telah di finetuning, mencapai akurasi 78.57% dan model ResNet50 yang dilatih menggunakan FER 2013 mencapai akurasi 71.42%. Penelitian ini berkontribusi pada pemahaman tentang transfer learning dalam pengenalan ekspresi wajah dan memberikan wawasan tentang performa arsitektur ResNet50 dan DenseNet121. Temuan studi ini memiliki implikasi untuk pengembangan sistem pengenalan ekspresi wajah yang lebih akurat, yang dapat diterapkan dalam berbagai bidang seperti pengenalan emosi, interaksi manusia-komputer, dan komputasi afektif.
English Abstract
Facial expression recognition is an important aspect of nonverbal communication, and its significance has triggered significant research in the field of artificial intelligence. This study investigates the effect of transfer learning using Convolutional Neural Network (CNN) architectures, specifically ResNet50 and DenseNet121, on facial expression recognition performance. The study aims to analyze the impact of transfer learning on their performance. The results show that implementing transfer learning on DenseNet121 and ResNet50 can significantly improve their performance, with DenseNet121 achieving an accuracy of 95.1% on the training set and 88.57% on the validation set, and ResNet50 achieving an accuracy of 100% on the training set and 81.43% on the validation set. The study also conducts testing by making predictions using a new dataset not included in the model training. The test results find that fine-tuning pre-trained models can improve model performance. The DenseNet121 model trained using FER-2013 and fine-tuned achieves an accuracy of 78.57%, and the ResNet50 model trained using FER-2013 achieves an accuracy of 71.42%. This research contributes to the understanding of transfer learning in facial expression recognition and provides insights into the performance of ResNet50 and DenseNet121 architectures. The findings of this study have implications for the development of more accurate facial expression recognition systems, which can be applied in various fields such as emotion recognition, human-computer interaction, and affective computing
| Item Type: | Thesis (Sarjana) |
|---|---|
| Identification Number: | 052415 |
| Uncontrolled Keywords: | pengenalan ekspresi wajah, transfer learning, fine tuning, Convolutional Neural Network (CNN), ResNet50, DenseNet121, FER-facial expression recognition, transfer learning, fine-tuning, Convolutional Neural Network (CNN), ResNet50, DenseNet121, FER-20132013.- |
| Divisions: | Fakultas Ilmu Komputer > Sistem Informasi |
| Depositing User: | Sugeng Moelyono |
| Date Deposited: | 04 Nov 2024 06:33 |
| Last Modified: | 04 Nov 2024 06:34 |
| URI: | http://repository.ub.ac.id/id/eprint/228038 |
![[thumbnail of DALAM MASA EMBARGO]](http://repository.ub.ac.id/style/images/fileicons/text.png) |
Text (DALAM MASA EMBARGO)
Geoffrey Edmund Pierre Purba.pdf Restricted to Registered users only Download (3MB) |
Actions (login required)
 |
View Item |