Deteksi Deepfake Real-Time pada Perangkat Mobile Menggunakan Arsitektur MobileViT-CBAM Teroptimasi
DOI:
https://doi.org/10.31004/riggs.v4i4.3634Keywords:
Deteksi Deepfake, MobileViT, Modified CBAM, Optimasi Model, Real-Time, Perangkat MobileAbstract
Perkembangan teknologi deepfake menghadirkan ancaman serius terhadap autentisitas informasi digital, terutama di media sosial dan konteks politik. Namun, sebagian besar model deteksi deepfake masih berukuran besar dan memiliki latensi tinggi, sehingga tidak efisien untuk dijalankan pada perangkat mobile. Penelitian ini mengusulkan arsitektur deteksi deepfake ringan berbasis MobileViT yang dipadukan dengan modified Convolutional Block Attention Module (CBAM) serta rangkaian optimasi model untuk memungkinkan inferensi real-time pada smartphone. MobileViT digunakan karena kemampuannya mengintegrasikan representasi lokal dan global secara efisien, sementara modified CBAM ditambahkan untuk meningkatkan fokus model pada area wajah yang sering dimanipulasi. Proses optimasi mencakup pruning 40%, quantization 8-bit, dan konversi TensorFlow Lite. Model dilatih menggunakan dataset FaceForensics++ dan Celeb-DF dengan total 84.690 frame yang diproses melalui MTCNN dan normalisasi 224×224 piksel. Hasil evaluasi menunjukkan bahwa model mencapai AUC 0.993 dan akurasi 96.4%, dengan ukuran akhir hanya 0.80 MB dan kecepatan 15.8 FPS pada perangkat simulasi MacBook M1 Pro. Ablation study mengonfirmasi kontribusi signifikan modified CBAM terhadap peningkatan performa, serta efektivitas quantization dalam menurunkan ukuran model tanpa mengorbankan akurasi. Temuan ini menunjukkan bahwa model MobileViT-CBAM teroptimasi mampu memberikan solusi deteksi deepfake yang akurat, ringan, dan dapat dijalankan secara real-time pada perangkat mobile tanpa ketergantungan cloud, sehingga berpotensi mendukung verifikasi konten multimedia dan mitigasi disinformasi di masyarakat.
Downloads
References
Cao, X., Zhong, P., Huang, Y., Huang, M., Huang, Z., Zou, T., & Xing, H. (2025). Research on Lightweight Algorithm Model for Precise Recognition and Detection of Outdoor Strawberries Based on Improved YOLOv5n. Agriculture, 15(1), 90. https://doi.org/10.3390/agriculture15010090
Duanmu, A., Xue, S., Li, Z., Zhang, Y., & Ni, C. (2025). Rep-MobileViT: Texture and Color Classification of Solid Wood Floors Based on a Re-Parameterized CNN-Transformer Hybrid Model. IEEE Access, 13, 39950–39963. https://doi.org/10.1109/ACCESS.2025.3545645
Jafar M Sidik. (2024, January 3). Ancaman “Deepfake”, pemanfaatan AI, dan Pemilu 2024. Aceh.Antaranews.Com.
Lee, J., Kwon, Y., Park, S., Yu, M., Park, J., & Song, H. (2024). Q-HyViT: Post-Training Quantization of Hybrid Vision Transformers With Bridge Block Reconstruction for IoT Systems. IEEE Internet of Things Journal, 11(22), 36384–36396. https://doi.org/10.1109/JIOT.2024.3403844
Liu, Y., Li, Y., Yi, X., Hu, Z., Zhang, H., & Liu, Y. (2022). Lightweight ViT Model for Micro-Expression Recognition Enhanced by Transfer Learning. Frontiers in Neurorobotics, 16. https://doi.org/10.3389/fnbot.2022.922761
Liu, Y., Xing, H., & Hou, T. (2025). Sea Surface Floating Small-Target Detection Based on Dual-Feature Images and Improved MobileViT. Journal of Marine Science and Engineering, 13(3), 572. https://doi.org/10.3390/jmse13030572
Palanisamy, B., Hassija, V., Chatterjee, A., Mandal, A., Chakraborty, D., Pandey, A., Chalapathi, G. S. S., & Kumar, D. (2025). Transformers for Vision: A Survey on Innovative Methods for Computer Vision. IEEE Access, 13, 95496–95523. https://doi.org/10.1109/ACCESS.2025.3571735
Tahyudin, I., Prabuwono, A. S., Dianingrum, M., Pandega, D. M., Winarto, E., Nazwan, Rozak, R. ’Abdul, Lestari, P., & Tikaningsih, A. (2024). ResNet-CBAM in Medical Imaging: A High-Accuracy Tool for Stroke Detection from CT Scans. 2024 8th International Conference on Information Technology, Information Systems and Electrical Engineering (ICITISEE), 551–556. https://doi.org/10.1109/ICITISEE63424.2024.10730079
Teh, S., Sivakumar, S., & Motalebi, F. (2024). Vision Transformers for Biomedical Applications *. 2024 International Conference on Green Energy, Computing and Sustainable Technology (GECOST), 195–201. https://doi.org/10.1109/GECOST60902.2024.10474871
Tomás Lujambio. (2023, July 20). Deepfakes Increase by 900% Annually: The Week in Cybersecurity. Mexicobusiness.News.
Zhang, G., Li, W., Tang, Y., Chen, S., & Wang, L. (2025). Lightweight CNN-ViT with cross-module representational constraint for express parcel detection. The Visual Computer, 41(5), 3283–3295. https://doi.org/10.1007/s00371-024-03602-0
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 Carli Apriansyah Hutagalung, Dinar Munggaran Akhmad, Ersa Resita, Talita Rahmi
This work is licensed under a Creative Commons Attribution 4.0 International License.
