DeepFlow-BiViTGAN: A Lightweight and Adaptive Traffic Detection System Combining GAN and Vision Transformer

With the continuous evolution of the network environment and attack patterns, existing deep learning-based traffic monitoring systems face challenges in terms of adaptability and computational resource requirements. To address these issues, this paper proposes an innovative traffic detection system, DeepFlow-BiViTGAN, which combines Generative Adversarial Network (GAN) and Vision Transformer (ViT) architectures with an improved loss function to enhance detection accuracy and system robustness. Experimental results indicate that DeepFlow-BiViTGAN achieves state-of-the-art detection performance on multiple public datasets when trained with approximately 3% to 5% of the total dataset. Its lightweight design enables efficient operation on resource-constrained IoT devices, offering excellent adaptability and scalability. This research provides new insights into the application of deep learning in traffic monitoring, particularly in IoT scenarios with limited data, and demonstrates significant advantages.