计算机科学
人工智能
数据挖掘
度量(数据仓库)
噪音(视频)
特征(语言学)
钥匙(锁)
机器学习
算法
人工神经网络
作者
Shraddha Shingne,Lajja Dadmal,Roshni Bajaj,Mahek Dhiman
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-04-01
卷期号:: 1-5
标识
DOI:10.1109/tqcebt67648.2026.11681629
摘要
The recent development of deepfakes has posed a serious challenge to the integrity and trustworthiness of digital media especially as online video platforms continue to spread at a remarkable rate. With the possibility to manipulate facial expressions, speech, and identities in the most realistic way, the deepfake technology presents acute threats to misinformation, identity fraud, personal, and national security. Since the deepfake generation methods have been constantly under development in terms of sophistication, the creation of correct and effective detection systems has become a more important issue. In this paper, a hybrid CNN-LSTM deep learning architecture is proposed, which is designed to enhance the accuracy of deep fake detection, and the computational efficiency of the deep fake detector framework should be sufficient to allow real-time applications. MobileNetV3 is used as the convolutional neural network (CNN) backbone in the framework to identify individual video frame discriminative spatial features. MobileNetV3 is selected because of its lightweight structure that makes it possible to implement it in resource-constrained settings. To capture temporal inconsistencies present across consecutive frames, the extracted spatial features are further processed using a Long Short-Term Memory (LSTM) network, allowing effective modeling of frame-to-frame dependencies.
科研通智能强力驱动
Strongly Powered by AbleSci AI