This paper presents a novel blockchain-based deepfake detection method that combines federated learning (FL) and deep learning (DL) models. The approach aims to address the growing threat of deepfake videos, which pose a significant challenge to media authenticity and security. By leveraging blockchain technology, the method ensures data source anonymity and privacy while maintaining the integrity of the training process. The core components of the BFLDL method include: 1. **Data Normalization**: A normalization technique is introduced to standardize data from diverse sources, ensuring uniformity in video quality and resolution. 2. **Feature Extraction**: The SegCaps and convolutional neural network (CNN) methods are used to enhance image feature extraction, followed by capsule network (CN) training to improve generalization. 3. **Federated Learning (FL)**: FL is employed to train a global model collaboratively, allowing multiple clients to contribute their locally trained models without sharing raw data. 4. **Transfer Learning (TL)**: TL is utilized to enhance the performance of the deep learning models by transferring learned features from large datasets to smaller datasets. 5. **Privacy and Security**: Blockchain technology is used to ensure data privacy and security during the training process, preventing data leakage and maintaining the confidentiality of source data. The effectiveness of the BFLDL method is validated through extensive experiments, demonstrating a significant improvement in accuracy and AUC metrics compared to benchmark models. The proposed approach not only enhances the detection of deepfake content but also addresses critical issues such as data privacy and security, making it a promising solution for advancing media authenticity and security in the digital age.This paper presents a novel blockchain-based deepfake detection method that combines federated learning (FL) and deep learning (DL) models. The approach aims to address the growing threat of deepfake videos, which pose a significant challenge to media authenticity and security. By leveraging blockchain technology, the method ensures data source anonymity and privacy while maintaining the integrity of the training process. The core components of the BFLDL method include: 1. **Data Normalization**: A normalization technique is introduced to standardize data from diverse sources, ensuring uniformity in video quality and resolution. 2. **Feature Extraction**: The SegCaps and convolutional neural network (CNN) methods are used to enhance image feature extraction, followed by capsule network (CN) training to improve generalization. 3. **Federated Learning (FL)**: FL is employed to train a global model collaboratively, allowing multiple clients to contribute their locally trained models without sharing raw data. 4. **Transfer Learning (TL)**: TL is utilized to enhance the performance of the deep learning models by transferring learned features from large datasets to smaller datasets. 5. **Privacy and Security**: Blockchain technology is used to ensure data privacy and security during the training process, preventing data leakage and maintaining the confidentiality of source data. The effectiveness of the BFLDL method is validated through extensive experiments, demonstrating a significant improvement in accuracy and AUC metrics compared to benchmark models. The proposed approach not only enhances the detection of deepfake content but also addresses critical issues such as data privacy and security, making it a promising solution for advancing media authenticity and security in the digital age.