Application of improved GAN-LSTM-based fake face detection technique in electronic data forensics

This study proposes a detection method based on an improved GAN-LSTM fusion model to address the challenges of insufficient feature extraction and poor robustness in face forgery detection during electronic data forensics. Core innovations include designing a composite LSTM module that integrates multiple LSTM units to enhance time feature processing capabilities. In addition, a relative average discriminator is introduced to optimize the balance of adversarial training. On the CelebA benchmark dataset, the model achieved a detection accuracy of 97.26% (significantly higher than the baseline model’s 92.14%), reduced inference time to 1.43 s, and reached an mean average precision (mAP) of 97.56%. During a 90-hour real-world simulation test, the model achieved a detection accuracy rate of 96.84%, with only 4.57% false positives. This performance surpassed that of mainstream comparison methods in terms of both computational efficiency and accuracy. This research demonstrates that the proposed method can efficiently identify complex forged faces. It provides a high-precision, low-cost solution for electronic data forensics, which is significant in preventing the abuse of forgery techniques. This approach provides a high-precision, low-computational-cost solution for forged facial recognition in digital forensics, demonstrating significant application value for safeguarding digital security.