TY  - JOUR
AU  - Akhtar, Farooq
AU  - Mahum, Rabbia
PY  - 2025
DA  - 2025/05/23
TI  - IRV2-hardswish Framework: A Deep Learning Approach for Deepfakes Detection and Classification
JO  - ICCK Journal of Image Analysis and Processing
T2  - ICCK Journal of Image Analysis and Processing
JF  - ICCK Journal of Image Analysis and Processing
VL  - 1
IS  - 2
SP  - 45
EP  - 56
DO  - 10.62762/JIAP.2025.421251
UR  - https://www.icck.org/article/abs/JIAP.2025.421251
KW  - deep learning
KW  - deepfakes
KW  - hardswish framework
KW  - DFDC dataset
AB  - Deep learning models are pivotal in the advancements of Artificial Intelligence (AI) due to rapid learning and decision-making across various fields such as healthcare, finance, and technology. However, a harmful utilization of deep learning models poses a threat to public welfare, national security, and confidentiality. One such example is Deepfakes, which creates and modifies audiovisual data that humans cannot tell apart from the real ones. Due to the progression of deep learning models that produce manipulated data, accurately detecting and classifying deepfake data becomes a challenge. This paper presents a groundbreaking IRV2-Hardswish Framework for deepfake detection, leveraging a hybrid deep learning architecture that synergizes residual blocks in CNNs and the Inception-Resnet-v2 model. By incorporating residual blocks to capture underlying audiovisual data layers and enhancing Inception-Resnet-v2 with Hardswish activation for robust feature extraction, our framework achieves accurate detection of deepfakes. Furthermore, additional dense layers are integrated to ensure precise classification, establishing a comprehensive and effective solution for deepfake detection. Further, a detailed comparison of our framework with the state-of-the-art CNN models reports that our framework outperforms with 98% accuracy, 96% precision, and 95% AUC using the Deep Fake Detection Challenge (DFDC) dataset. The DFDC dataset is the largest, consisting of approximately 5,000 clips, including 1,132 actual and 4,118 false ones. The results report the efficiency of the proposed framework. These results demonstrate the framework's effectiveness in deepfake detection.
SN  - 3068-6679
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Akhtar2025IRV2hardsw,
  author = {Farooq Akhtar and Rabbia Mahum},
  title = {IRV2-hardswish Framework: A Deep Learning Approach for Deepfakes Detection and Classification},
  journal = {ICCK Journal of Image Analysis and Processing},
  year = {2025},
  volume = {1},
  number = {2},
  pages = {45-56},
  doi = {10.62762/JIAP.2025.421251},
  url = {https://www.icck.org/article/abs/JIAP.2025.421251},
  abstract = {Deep learning models are pivotal in the advancements of Artificial Intelligence (AI) due to rapid learning and decision-making across various fields such as healthcare, finance, and technology. However, a harmful utilization of deep learning models poses a threat to public welfare, national security, and confidentiality. One such example is Deepfakes, which creates and modifies audiovisual data that humans cannot tell apart from the real ones. Due to the progression of deep learning models that produce manipulated data, accurately detecting and classifying deepfake data becomes a challenge. This paper presents a groundbreaking IRV2-Hardswish Framework for deepfake detection, leveraging a hybrid deep learning architecture that synergizes residual blocks in CNNs and the Inception-Resnet-v2 model. By incorporating residual blocks to capture underlying audiovisual data layers and enhancing Inception-Resnet-v2 with Hardswish activation for robust feature extraction, our framework achieves accurate detection of deepfakes. Furthermore, additional dense layers are integrated to ensure precise classification, establishing a comprehensive and effective solution for deepfake detection. Further, a detailed comparison of our framework with the state-of-the-art CNN models reports that our framework outperforms with 98\% accuracy, 96\% precision, and 95\% AUC using the Deep Fake Detection Challenge (DFDC) dataset. The DFDC dataset is the largest, consisting of approximately 5,000 clips, including 1,132 actual and 4,118 false ones. The results report the efficiency of the proposed framework. These results demonstrate the framework's effectiveness in deepfake detection.},
  keywords = {deep learning, deepfakes, hardswish framework, DFDC dataset},
  issn = {3068-6679},
  publisher = {Institute of Central Computation and Knowledge}
}