Rangga Kalam Sidiq, - (2024) DETEKSI EMOSI DAN AROUSAL-VALENCE DALAM APPLICATION PROGRAMMING INTERFACE MENGGUNAKAN HAAR CASCADE DAN EFFICIENTNET PADA DATASET AFFECTNET. S1 thesis, Universitas Pendidikan Indonesia.
Abstract
Deteksi emosi dan arousal-valence masih menjadi topik penting dalam pengembangan sistem pengenalan emosi, serta kebutuhan untuk meningkatkan kinerja model tetap menjadi tantangan. Saat ini terdapat beberapa sistem pengenalan emosi yang masih belum menerapkan deteksi wajah menggunakan Haar Cascade baik dalam pra-proses data untuk pelatihan maupun proses inferensi dengan integrasi model yang memberikan keluaran untuk klasifikasi emosi sekaligus dimensi valensi dan arousal. Selain itu, penerapan model pengenalan emosi ke dalam bentuk API yang diterapkan pada layanan penyimpanan awan juga masih jarang. Penelitian ini bertujuan untuk menerapkan model deteksi dengan tiga keluaran berisikan klasifikasi emosi, nilai valensi, dan nilai arousal menggunakan EfficientNet serta Haar Cascade yang bertujuan untuk melakukan deteksi wajah pada proses pelatihan dan inferensi model. Dari beberapa eksperimen yang dilakukan meliputi penerapan beberapa metode pembelajaran transfer seperti pembekuan dan pencairan lapisan model EfficientNetB0 dan EfficientNetB7, penyetelan model dengan tingkat pembelajaran yang berbeda, dan penerapan augmentasi gambar mendapatkan hasil model terbaik dengan akurasi pada keluaran klasifikasi emosi sebesar 56%, pada keluaran nilai valensi MAE 0.2889, MSE 0.1333, RMSE 0.3651, dan pada keluaran nilai arousal MAE 0.2795, MSE 0.1244, RMSE 0.3527, yang menunjukkan nilai akurasi dan kesalahan yang cukup baik. Hasil model terbaik selanjutnya dilakukan implementasi pada API dan diterapkan pada layanan Google Cloud dengan hasil akurasi 99.55% akurat sesuai dengan hasil sebelum diimplementasi dengan hasil rata-rata waktu request 281.61 milidetik untuk 28.1 request per detik dari total 35 pengguna virtual. ----------- Emotion and arousal-valence detection are still important topics in the development of emotion recognition systems, and the need for improvement in the performance of emotion and engagement detection models is still a challenge. Currently, there are several emotion recognition systems that have not yet implemented Haar Cascade for face detection, either in the pre-processing of data for training, or in the inference process with the integration of models that provide output for emotion classification as well as valence and arousal dimensions. Furthermore, the application of emotion recognition models in the form of APIs applied to cloud storage services is still rare. This research aims to implement a detection model with three outputs including emotion classification, valence and arousal value using EfficientNet and Haar Cascade, which aims to perform face detection in the model training and inference process. From several experiments conducted, including the application of several transfer learning methods such as freezing and fine-tuning the model layers of EfficientNetB0 and EfficientNetB7, tuning the model with different learning levels, and applying image augmentation, the best model results with accuracy at the output of emotion classification of 56%, at the output of valence value MAE 0.2889, MSE 0.1333, RMSE 0.3651, and at the output of arousal value MAE 0.2795, MSE 0.1244, RMSE 0.3527, which shows quite good accuracy and error values. The best model results are then implemented on the API and applied to the Google Cloud service with 99.55% accuracy according to the results before implementation, with an average request time of 281.61 milliseconds for 28.1 requests per second from a total of 35 virtual users.
![[img]](http://repository.upi.edu/style/images/fileicons/text.png) |
Text
S_RPL_2008752_Title.pdf Download (4MB) |
|
Text
S_RPL_2008752_Chapter1.pdf Download (2MB) |
|
Text
S_RPL_2008752_Chapter2.pdf Restricted to Staf Perpustakaan Download (5MB) | Request a copy |
|
Text
S_RPL_2008752_Chapter3.pdf Download (2MB) |
|
Text
S_RPL_2008752_Chapter4.pdf Restricted to Staf Perpustakaan Download (14MB) | Request a copy |
|
Text
S_RPL_2008752_Chapter5.pdf Download (708kB) |
|
Text
S_RPL_2008752_Appendix.pdf Restricted to Staf Perpustakaan Download (1MB) | Request a copy |
| Item Type: | Thesis (S1) |
|---|---|
| Additional Information: | https://scholar.google.com/citations?view_op=new_profile&hl=en&authuser=3 ID SINTA Dosen Pembimbing: Dian Anggraini: 6681986 Asyifa Imanda Septiana: 6681802 |
| Uncontrolled Keywords: | Deteksi Emosi dan Arousal-Valence, Valensi dan Arousal, EfficientNet, AffectNet, Haar Cascade |
| Subjects: | Q Science > QA Mathematics Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software T Technology > T Technology (General) |
| Divisions: | UPI Kampus cibiru > S1 Rekayasa Perangkaat Lunak |
| Depositing User: | Rangga Kalam Sidiq |
| Date Deposited: | 11 Sep 2024 02:21 |
| Last Modified: | 11 Sep 2024 02:21 |
| URI: | http://repository.upi.edu/id/eprint/121784 |
Actions (login required)
 |
View Item |